jjrussell10 commited on
Commit
c9656c7
·
verified ·
1 Parent(s): b080c5f

Model cards: cite russell2026wildai; root tokenizer (part 4)

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. 86.2m-g041-ai-r4.22/README.md +2 -2
  2. 86.2m-g041-ai-r8/README.md +2 -2
  3. 86.2m-g041-control/README.md +2 -2
  4. 86.2m-g041-human-r2/README.md +2 -2
  5. 86.2m-g042-ai-r1/README.md +2 -2
  6. 86.2m-g042-control/README.md +2 -2
  7. 86.2m-g043-ai-r1/README.md +2 -2
  8. 86.2m-g043-control/README.md +2 -2
  9. 86.2m-g044-ai-r0.025/README.md +2 -2
  10. 86.2m-g044-ai-r0.05/README.md +2 -2
  11. 86.2m-g044-ai-r0.1/README.md +2 -2
  12. 86.2m-g044-ai-r0.25/README.md +2 -2
  13. 86.2m-g044-ai-r0.5/README.md +2 -2
  14. 86.2m-g044-ai-r1/README.md +2 -2
  15. 86.2m-g044-ai-r2/README.md +2 -2
  16. 86.2m-g044-ai-r4/README.md +2 -2
  17. 86.2m-g044-ai-r8/README.md +2 -2
  18. 86.2m-g044-control/README.md +2 -2
  19. 86.2m-g044-human-r0.025/README.md +2 -2
  20. 86.2m-g044-human-r0.05/README.md +2 -2
  21. 86.2m-g044-human-r2/README.md +2 -2
  22. 86.2m-g045-ai-r0.05/README.md +2 -2
  23. 86.2m-g045-ai-r0.1/README.md +2 -2
  24. 86.2m-g045-ai-r0.25/README.md +2 -2
  25. 86.2m-g045-ai-r0.5/README.md +2 -2
  26. 86.2m-g045-ai-r1/README.md +2 -2
  27. 86.2m-g045-control/README.md +2 -2
  28. 86.2m-g046-ai-r0.05/README.md +2 -2
  29. 86.2m-g046-ai-r0.1/README.md +2 -2
  30. 86.2m-g046-ai-r0.25/README.md +2 -2
  31. 86.2m-g046-ai-r0.5/README.md +2 -2
  32. 86.2m-g046-ai-r1/README.md +2 -2
  33. 86.2m-g046-control/README.md +2 -2
  34. 86.2m-g047-ai-r0.05/README.md +2 -2
  35. 86.2m-g047-ai-r0.1/README.md +2 -2
  36. 86.2m-g047-ai-r0.25/README.md +2 -2
  37. 86.2m-g047-ai-r0.5/README.md +2 -2
  38. 86.2m-g047-ai-r1/README.md +2 -2
  39. 86.2m-g047-ai-r2.07/README.md +2 -2
  40. 86.2m-g047-ai-r4.2/README.md +2 -2
  41. 86.2m-g047-control/README.md +2 -2
  42. 86.2m-g048-ai-r0.25/README.md +2 -2
  43. 86.2m-g048-ai-r1/README.md +2 -2
  44. 86.2m-g048-control/README.md +2 -2
  45. 86.2m-g049-ai-r0.05/README.md +2 -2
  46. 86.2m-g049-ai-r0.1/README.md +2 -2
  47. 86.2m-g049-ai-r0.25/README.md +2 -2
  48. 86.2m-g049-ai-r0.5/README.md +2 -2
  49. 86.2m-g049-ai-r1/README.md +2 -2
  50. 86.2m-g049-ai-r2/README.md +2 -2
86.2m-g041-ai-r4.22/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g041-ai-r8/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g041-control/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g041-human-r2/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g042-ai-r1/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g042-control/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g043-ai-r1/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g043-control/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g044-ai-r0.025/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g044-ai-r0.05/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g044-ai-r0.1/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g044-ai-r0.25/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g044-ai-r0.5/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g044-ai-r1/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g044-ai-r2/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g044-ai-r4/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g044-ai-r8/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g044-control/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g044-human-r0.025/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g044-human-r0.05/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g044-human-r2/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g045-ai-r0.05/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g045-ai-r0.1/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g045-ai-r0.25/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g045-ai-r0.5/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g045-ai-r1/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g045-control/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g046-ai-r0.05/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g046-ai-r0.1/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g046-ai-r0.25/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g046-ai-r0.5/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g046-ai-r1/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g046-control/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g047-ai-r0.05/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g047-ai-r0.1/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g047-ai-r0.25/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g047-ai-r0.5/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g047-ai-r1/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g047-ai-r2.07/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g047-ai-r4.2/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g047-control/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g048-ai-r0.25/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g048-ai-r1/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g048-control/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g049-ai-r0.05/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g049-ai-r0.1/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g049-ai-r0.25/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g049-ai-r0.5/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g049-ai-r1/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
86.2m-g049-ai-r2/README.md CHANGED
@@ -18,7 +18,7 @@ tags:
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
- A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
@@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que
95
  ## Citation
96
 
97
  ```bibtex
98
- @article{russell2026aitoken,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},
 
18
 
19
  > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license.
20
 
21
+ A 86.2M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training.
22
 
23
  ## Training data
24
 
 
95
  ## Citation
96
 
97
  ```bibtex
98
+ @article{russell2026wildai,
99
  title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text},
100
  author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley},
101
  journal = {arXiv preprint arXiv:2609.40295},