Datasets:

SupraLabs
/

Supra-microBench

---
language:
- en
license: other
pretty_name: Supra-μBench
tags:
- benchmark
- language-model-evaluation
size_categories:
- 1K<n<10K
---
<h1 align="center">Supra-μBench • 512 max seq len • Every small model welcome</h1>

![banner](https://cdn-uploads.huggingface.co/production/uploads/697f2832c2c5e4daa93cece7/KIfNJ1k8sx7F4XRkpZOjt.png)

Supra-μBench (Supra-microBench) v0.2.0 is a log-likelihood benchmark for very small base language models, built by SupraLabs. It's built for evaluation your small models under 100M parameters (yea, we evaluated Qwen3-0.6B too but that was only for comparison!). A model is ranked by how much probability it assigns to the correct continuation (loglikelihood like you know it from the LM-Eval tool).

The set has about 1060 English items across 18 tasks and six domains. Items are produced by seeded generators plus a smaller set of hand-written items. The generation seed is 1234. Each task has its own random stream, so editing one task does not reshuffle the others.

## IMPORTANT

1. Do NOT train on this dataset! It's for evaluation only!!
2. This is for BASE models only - not for instruction tuned models. Maybe we'll release a benchmark for instruct models too...
3. Your model SHOULD have at least 512 tokens of context window

## Domains & Tasks

The μBench Intelligence Index is a weighted mean of the six domain scores. Each domain score is the unweighted mean of its task scores. Every score is on a 0 to 100 scale.

| Domain | Name | Weight |
| --- | --- | --- |
| D1 | Linguistic Competence | 0.15 |
| D2 | Lexicon & Semantics | 0.15 |
| D3 | World Knowledge | 0.15 |
| D4 | Context Use & Reading | 0.20 |
| D5 | Elementary Reasoning | 0.20 |
| D6 | Coherence, Common Sense & LM | 0.15 |

## How a choice item is scored

The evaluator loads a causal language model and its tokenizer. Every sequence starts with a BOS token, or EOS if the tokenizer has no BOS. Continuations are scored by summed log-probability. No tokens are sampled.

Three reductions are used. `sum` compares raw log-probabilities. `mean_byte` divides by the UTF-8 length of the continuation so a longer string is not punished. `pmi` subtracts the log-probability of the same continuation under a null context, so a word that is likely in general does not win just for being frequent.

The predicted choice is the one with the highest score. Accuracy is then chance-normalized:

**score = 100 * max(0, (accuracy - chance) / (1 - chance))**

A model at random chance scores 0. A model that is always correct scores 100. Scores are not allowed to go below 0 (and what bad model would that be 😭🤣).

On tasks marked with group accuracy, a contrast pair is one unit. Both members have to  be correct. That cancels a model that always prefers one surface form. Units that have no group are scored alone.

The bits-per-byte task is separate. Fresh text is scored with a sliding window. Bits per byte are mapped onto 0 to 100 with provisional anchors: a floor of 2.5 (a weak unigram-like model) and a ceiling of 0.5 (a strong big LLM like a 2B or 4B model). The mapping is logarithmic and clipped to 0 to 100.

Reported standard errors are item-sampling errors. Treat an index gap as noise when it is smaller than 2 times the square root of the sum of the two squared standard errors.

## How to submit a run

Run the SupraLabs evaluator on a local model folder or a Hub model id. The folder must contain the tokenizer.

That JSON is the leaderboard source. It stores the model path, the parameter count, per-task scores, the six domain scores, and `mubench_intelligence_index`.

You can find `evaluate.py` and `tasks.py` in the files list of the repo.

Example of how to run the evaluation:
```bash
python3 evaluate.py --model_dir SupraLabs/SupraNeo-4M --trust_remote_code --show_levels --output SupraNeo4M-Eval-Results.json
```

## Leaderboard

One run is on file. Rank is by INDEX, high to low.

| Rank | Modelname | Org/Creator | Params | INDEX | D1 | D2 | D3 | D4 | D5 | D6 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 🥇 1 | Qwen3-0.6B | Qwen | 596.0M | 69.82 | 59.82 | 79.42 | 87.94 | 64.03 | 77.10 | 50.10 |
| 🥈 2 | cagliostro-v3 | bench-labs | 146.4M | 60.16 | 56.99 | 86.94 | 72.74 | 49.99 | 51.09 | 49.65 |
| 🥉 3 | SmolLM2-135M-Base | HuggingFaceTB | 134.5M | 58.90 | 62.18 | 86.33 | 86.85 | 40.85 | 45.93 | 41.62 |
| 4 | cagliostro-v3.5 | bench-labs | 146.4M | 57.51 | 53.84 | 84.43 | 70.69 | 46.56 | 50.33 | 45.23 |
| 5 | GPT-X2.5-135M | AxiomicLabs | 135.0M | 56.18 | 60.91 | 82.78 | 75.28 | 39.29 | 44.67 | 43.64 |
| 6 | GPT-X3-150M | AxiomicLabs | 149.4M | 55.04 | 47.4 | 78.7 | 71.0 | 42.3 | 54.4 | 41.0 |
| 7 | SmolLM-135M | HuggingFaceTB | 134.5M | 54.86 | 53.19 | 84.51 | 78.29 | 40.00 | 37.99 | 45.73 |
| 8 | tinctura-v1 | bench-labs | 96.2M | 53.99 | 53.62 | 82.82 | 67.08 | 40.52 | 45.09 | 42.25 |
| 9 | GPT-X2-125M | AxiomicLabs | 125.1M | 53.15 | 54.18 | 82.82 | 66.72 | 38.26 | 39.31 | 47.19 |
| 10 | ZeroS-v0.1-150M | fromziro | 151.6M | 49.88 | 65.13 | 72.69 | 60.57 | 32.48 | 34.55 | 44.76 |
| 11 | JugnuLM-110M-R2plus | altslate | 109.7M | 49.62 | 64.61 | 73.54 | 65.52 | 36.61 | 29.99 | 38.33 |
| 12 | BananaMind-2-Pro | BananaMind | 139.0M | 48.73 | 47.88 | 75.81 | 73.23 | 35.41 | 26.75 | 45.07 |
| 13 | Speck2-140M-Instruct | specklabs | 140.7M | 47.75 | 54.73 | 75.23 | 56.22 | 36.61 | 33.29 | 38.98 |
| 14 | Rose-1.5-Medium | GODELEV | 98.6M | 47.33 | 56.87 | 75.07 | 60.68 | 31.42 | 28.02 | 43.65 |
| 15 | cRia-LM-75M | sz14 | 75.7M | 46.60 | 57.19 | 80.24 | 65.17 | 24.14 | 32.96 | 31.94 |
| 16 | void.1 | dotlabs | 90.1M | 46.45 | 54.46 | 71.12 | 52.01 | 31.91 | 40.70 | 35.25 |
| 17 | Supra2-100M | SupraLabs | 100.7M | 45.38 | 58.14 | 73.42 | 64.67 | 26.62 | 21.75 | 41.77 |
| 18 | Haidass1.5-143M | DALabCommunity | 143.1M | 43.74 | 40.69 | 78.47 | 59.61 | 29.38 | 28.34 | 35.88 |
| 19 | Flare | SurjoLabs | 130.4M | 43.62 | 37.17 | 69.03 | 72.98 | 28.14 | 27.49 | 37.43 |
| 20 | Pythia-160M | EleutherAI | 162.3M | 42.47 | 58.69 | 72.57 | 48.51 | 25.48 | 26.79 | 33.69 |
| 21 | Rose-Pro | GODELEV | 151.3M | 42.04 | 32.53 | 75.97 | 57.43 | 32.67 | 27.67 | 33.92 |
| 22 | Supra-50M-Base | SupraLabs | 51.8M | 40.11 | 45.11 | 75.34 | 60.57 | 30.20 | 12.54 | 29.40 |
| 23 | Surjo-50m | SurjoLabs | 53.8M | 38.83 | 43.13 | 76.84 | 42.01 | 29.38 | 22.80 | 27.34 |
| 24 | Surjo-100m | SurjoLabs | 97.7M | 37.26 | 25.35 | 68.33 | 55.16 | 29.10 | 19.63 | 34.58 |
| 25 | Rambley-150M-Base | allura-org | 149.9M | 33.64 | 36.97 | 56.20 | 42.99 | 21.82 | 21.38 | 30.50 |
| 26 | Supra2-Medium-Base | SupraLabs | 25.4M | 26.70 | 0.44 | 57.07 | 49.50 | 19.50 | 14.35 | 25.84 |
| 27 | Pythia-70M | EleutherAI | 70.4M | 25.45 | 24.57 | 48.26 | 34.54 | 14.69 | 18.18 | 18.47 |
| 28 | ForgePlex-M2-9M | ForgeWorks | 9.9M | 23.78 | 31.47 | 34.27 | 22.97 | 22.38 | 15.08 | 19.85 |
| 29 | BananaMind-2-Nano | BananaMind | 10.0M | 21.51 | 11.25 | 46.03 | 27.90 | 17.31 | 14.96 | 15.21 |
| 30 | NCN-RW | AxiomicLabs | 9.7M | 18.99 | 16.26 | 33.53 | 18.52 | 18.13 | 12.63 | 17.27 |
| 31 | Spark-2A | SurjoLabs | 9.1M | 18.86 | 13.96 | 42.59 | 15.97 | 16.46 | 12.80 | 14.20 |
| 32 | Pythia-14M | EleutherAI | 14.1M | 15.10 | 4.00 | 33.19 | 23.44 | 4.84 | 13.80 | 15.19 |
| 33 | StorySupra-10M | SupraLabs | 12.6M | 14.46 | 4.36 | 13.90 | 0.80 | 34.15 | 5.15 | 24.97 |
| 34 | SupraNeo-4M | SupraLabs | 4.1M | 14.31 | 3.33 | 29.06 | 19.72 | 14.70 | 8.66 | 12.14 |
| 35 | pulvis-v2 | bench-labs | 3.0M | 13.94 | 3.11 | 33.19 | 6.10 | 12.22 | 16.27 | 12.52 |
| 36 | BananaMind-2-Micro | BananaMind | 2.9M | 12.04 | 7.35 | 16.58 | 3.81 | 15.89 | 14.29 | 12.25 |
| 37 | Supra-Mini-v5-8M-FINAL | SupraLabs | 7.9M | 11.95 | 0.00 | 28.17 | 19.50 | 9.22 | 5.86 | 11.85 |
| 38 | Tokle-3M | techdotus | 2.9M | 11.89 | 0.00 | 24.05 | 9.36 | 10.32 | 10.50 | 18.07 |
| 39 | Supra-Mini-v3-0.5M | SupraLabs | 467648 | 3.05 | 0.00 | 3.70 | 0.72 | 1.55 | 5.88 | 5.98 |

> Note: we did NOT rank every model that is ranked on the [Open SLM Leaderboard](https://huggingface.co/spaces/AxiomicLabs/Open_SLM_Leaderboard). If you have new ranks, you can post them in your model card.

## Intended use

Use the index to compare small base models on grammar, lexicon, facts, reading, elementary reasoning, and local coherence. Do not use it as a chat, instruction-following, or safety score. DO NOT TRAIN ON THE TASKS!!!

## Limitations

Language: English only. Several tasks have only a few units, so standard errors are sometimes wide. World-knowledge items reflect widely taught facts, not a neutral sample of the world. The bits-per-byte anchors are provisional. A model whose context window is shorter than an item is left-truncated, which can hurt the reading tasks but we tried to keep the tasks inside a 512 token window.

## Credits

Supra-μBench was built by SupraLabs. Version 0.2.0.

	---
	language:
	- en
	license: other
	pretty_name: Supra-μBench
	tags:
	- benchmark
	- language-model-evaluation
	size_categories:
	- 1K<n<10K
	---

	<h1 align="center">Supra-μBench • 512 max seq len • Every small model welcome</h1>

	![banner](https://cdn-uploads.huggingface.co/production/uploads/697f2832c2c5e4daa93cece7/KIfNJ1k8sx7F4XRkpZOjt.png)

	Supra-μBench (Supra-microBench) v0.2.0 is a log-likelihood benchmark for very small base language models, built by SupraLabs. It's built for evaluation your small models under 100M parameters (yea, we evaluated Qwen3-0.6B too but that was only for comparison!). A model is ranked by how much probability it assigns to the correct continuation (loglikelihood like you know it from the LM-Eval tool).

	The set has about 1060 English items across 18 tasks and six domains. Items are produced by seeded generators plus a smaller set of hand-written items. The generation seed is 1234. Each task has its own random stream, so editing one task does not reshuffle the others.

	## IMPORTANT

	1. Do NOT train on this dataset! It's for evaluation only!!
	2. This is for BASE models only - not for instruction tuned models. Maybe we'll release a benchmark for instruct models too...
	3. Your model SHOULD have at least 512 tokens of context window

	## Domains & Tasks

	The μBench Intelligence Index is a weighted mean of the six domain scores. Each domain score is the unweighted mean of its task scores. Every score is on a 0 to 100 scale.

	\| Domain \| Name \| Weight \|
	\| --- \| --- \| --- \|
	\| D1 \| Linguistic Competence \| 0.15 \|
	\| D2 \| Lexicon & Semantics \| 0.15 \|
	\| D3 \| World Knowledge \| 0.15 \|
	\| D4 \| Context Use & Reading \| 0.20 \|
	\| D5 \| Elementary Reasoning \| 0.20 \|
	\| D6 \| Coherence, Common Sense & LM \| 0.15 \|

	## How a choice item is scored

	The evaluator loads a causal language model and its tokenizer. Every sequence starts with a BOS token, or EOS if the tokenizer has no BOS. Continuations are scored by summed log-probability. No tokens are sampled.

	Three reductions are used. `sum` compares raw log-probabilities. `mean_byte` divides by the UTF-8 length of the continuation so a longer string is not punished. `pmi` subtracts the log-probability of the same continuation under a null context, so a word that is likely in general does not win just for being frequent.

	The predicted choice is the one with the highest score. Accuracy is then chance-normalized:

	*score = 100 max(0, (accuracy - chance) / (1 - chance))**

	A model at random chance scores 0. A model that is always correct scores 100. Scores are not allowed to go below 0 (and what bad model would that be 😭🤣).

	On tasks marked with group accuracy, a contrast pair is one unit. Both members have to be correct. That cancels a model that always prefers one surface form. Units that have no group are scored alone.

	The bits-per-byte task is separate. Fresh text is scored with a sliding window. Bits per byte are mapped onto 0 to 100 with provisional anchors: a floor of 2.5 (a weak unigram-like model) and a ceiling of 0.5 (a strong big LLM like a 2B or 4B model). The mapping is logarithmic and clipped to 0 to 100.

	Reported standard errors are item-sampling errors. Treat an index gap as noise when it is smaller than 2 times the square root of the sum of the two squared standard errors.

	## How to submit a run

	Run the SupraLabs evaluator on a local model folder or a Hub model id. The folder must contain the tokenizer.

	That JSON is the leaderboard source. It stores the model path, the parameter count, per-task scores, the six domain scores, and `mubench_intelligence_index`.

	You can find `evaluate.py` and `tasks.py` in the files list of the repo.

	Example of how to run the evaluation:
	```bash
	python3 evaluate.py --model_dir SupraLabs/SupraNeo-4M --trust_remote_code --show_levels --output SupraNeo4M-Eval-Results.json
	```

	## Leaderboard

	One run is on file. Rank is by INDEX, high to low.

	\| Rank \| Modelname \| Org/Creator \| Params \| INDEX \| D1 \| D2 \| D3 \| D4 \| D5 \| D6 \|
	\| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \|
	\| 🥇 1 \| Qwen3-0.6B \| Qwen \| 596.0M \| 69.82 \| 59.82 \| 79.42 \| 87.94 \| 64.03 \| 77.10 \| 50.10 \|
	\| 🥈 2 \| cagliostro-v3 \| bench-labs \| 146.4M \| 60.16 \| 56.99 \| 86.94 \| 72.74 \| 49.99 \| 51.09 \| 49.65 \|
	\| 🥉 3 \| SmolLM2-135M-Base \| HuggingFaceTB \| 134.5M \| 58.90 \| 62.18 \| 86.33 \| 86.85 \| 40.85 \| 45.93 \| 41.62 \|
	\| 4 \| cagliostro-v3.5 \| bench-labs \| 146.4M \| 57.51 \| 53.84 \| 84.43 \| 70.69 \| 46.56 \| 50.33 \| 45.23 \|
	\| 5 \| GPT-X2.5-135M \| AxiomicLabs \| 135.0M \| 56.18 \| 60.91 \| 82.78 \| 75.28 \| 39.29 \| 44.67 \| 43.64 \|
	\| 6 \| GPT-X3-150M \| AxiomicLabs \| 149.4M \| 55.04 \| 47.4 \| 78.7 \| 71.0 \| 42.3 \| 54.4 \| 41.0 \|
	\| 7 \| SmolLM-135M \| HuggingFaceTB \| 134.5M \| 54.86 \| 53.19 \| 84.51 \| 78.29 \| 40.00 \| 37.99 \| 45.73 \|
	\| 8 \| tinctura-v1 \| bench-labs \| 96.2M \| 53.99 \| 53.62 \| 82.82 \| 67.08 \| 40.52 \| 45.09 \| 42.25 \|
	\| 9 \| GPT-X2-125M \| AxiomicLabs \| 125.1M \| 53.15 \| 54.18 \| 82.82 \| 66.72 \| 38.26 \| 39.31 \| 47.19 \|
	\| 10 \| ZeroS-v0.1-150M \| fromziro \| 151.6M \| 49.88 \| 65.13 \| 72.69 \| 60.57 \| 32.48 \| 34.55 \| 44.76 \|
	\| 11 \| JugnuLM-110M-R2plus \| altslate \| 109.7M \| 49.62 \| 64.61 \| 73.54 \| 65.52 \| 36.61 \| 29.99 \| 38.33 \|
	\| 12 \| BananaMind-2-Pro \| BananaMind \| 139.0M \| 48.73 \| 47.88 \| 75.81 \| 73.23 \| 35.41 \| 26.75 \| 45.07 \|
	\| 13 \| Speck2-140M-Instruct \| specklabs \| 140.7M \| 47.75 \| 54.73 \| 75.23 \| 56.22 \| 36.61 \| 33.29 \| 38.98 \|
	\| 14 \| Rose-1.5-Medium \| GODELEV \| 98.6M \| 47.33 \| 56.87 \| 75.07 \| 60.68 \| 31.42 \| 28.02 \| 43.65 \|
	\| 15 \| cRia-LM-75M \| sz14 \| 75.7M \| 46.60 \| 57.19 \| 80.24 \| 65.17 \| 24.14 \| 32.96 \| 31.94 \|
	\| 16 \| void.1 \| dotlabs \| 90.1M \| 46.45 \| 54.46 \| 71.12 \| 52.01 \| 31.91 \| 40.70 \| 35.25 \|
	\| 17 \| Supra2-100M \| SupraLabs \| 100.7M \| 45.38 \| 58.14 \| 73.42 \| 64.67 \| 26.62 \| 21.75 \| 41.77 \|
	\| 18 \| Haidass1.5-143M \| DALabCommunity \| 143.1M \| 43.74 \| 40.69 \| 78.47 \| 59.61 \| 29.38 \| 28.34 \| 35.88 \|
	\| 19 \| Flare \| SurjoLabs \| 130.4M \| 43.62 \| 37.17 \| 69.03 \| 72.98 \| 28.14 \| 27.49 \| 37.43 \|
	\| 20 \| Pythia-160M \| EleutherAI \| 162.3M \| 42.47 \| 58.69 \| 72.57 \| 48.51 \| 25.48 \| 26.79 \| 33.69 \|
	\| 21 \| Rose-Pro \| GODELEV \| 151.3M \| 42.04 \| 32.53 \| 75.97 \| 57.43 \| 32.67 \| 27.67 \| 33.92 \|
	\| 22 \| Supra-50M-Base \| SupraLabs \| 51.8M \| 40.11 \| 45.11 \| 75.34 \| 60.57 \| 30.20 \| 12.54 \| 29.40 \|
	\| 23 \| Surjo-50m \| SurjoLabs \| 53.8M \| 38.83 \| 43.13 \| 76.84 \| 42.01 \| 29.38 \| 22.80 \| 27.34 \|
	\| 24 \| Surjo-100m \| SurjoLabs \| 97.7M \| 37.26 \| 25.35 \| 68.33 \| 55.16 \| 29.10 \| 19.63 \| 34.58 \|
	\| 25 \| Rambley-150M-Base \| allura-org \| 149.9M \| 33.64 \| 36.97 \| 56.20 \| 42.99 \| 21.82 \| 21.38 \| 30.50 \|
	\| 26 \| Supra2-Medium-Base \| SupraLabs \| 25.4M \| 26.70 \| 0.44 \| 57.07 \| 49.50 \| 19.50 \| 14.35 \| 25.84 \|
	\| 27 \| Pythia-70M \| EleutherAI \| 70.4M \| 25.45 \| 24.57 \| 48.26 \| 34.54 \| 14.69 \| 18.18 \| 18.47 \|
	\| 28 \| ForgePlex-M2-9M \| ForgeWorks \| 9.9M \| 23.78 \| 31.47 \| 34.27 \| 22.97 \| 22.38 \| 15.08 \| 19.85 \|
	\| 29 \| BananaMind-2-Nano \| BananaMind \| 10.0M \| 21.51 \| 11.25 \| 46.03 \| 27.90 \| 17.31 \| 14.96 \| 15.21 \|
	\| 30 \| NCN-RW \| AxiomicLabs \| 9.7M \| 18.99 \| 16.26 \| 33.53 \| 18.52 \| 18.13 \| 12.63 \| 17.27 \|
	\| 31 \| Spark-2A \| SurjoLabs \| 9.1M \| 18.86 \| 13.96 \| 42.59 \| 15.97 \| 16.46 \| 12.80 \| 14.20 \|
	\| 32 \| Pythia-14M \| EleutherAI \| 14.1M \| 15.10 \| 4.00 \| 33.19 \| 23.44 \| 4.84 \| 13.80 \| 15.19 \|
	\| 33 \| StorySupra-10M \| SupraLabs \| 12.6M \| 14.46 \| 4.36 \| 13.90 \| 0.80 \| 34.15 \| 5.15 \| 24.97 \|
	\| 34 \| SupraNeo-4M \| SupraLabs \| 4.1M \| 14.31 \| 3.33 \| 29.06 \| 19.72 \| 14.70 \| 8.66 \| 12.14 \|
	\| 35 \| pulvis-v2 \| bench-labs \| 3.0M \| 13.94 \| 3.11 \| 33.19 \| 6.10 \| 12.22 \| 16.27 \| 12.52 \|
	\| 36 \| BananaMind-2-Micro \| BananaMind \| 2.9M \| 12.04 \| 7.35 \| 16.58 \| 3.81 \| 15.89 \| 14.29 \| 12.25 \|
	\| 37 \| Supra-Mini-v5-8M-FINAL \| SupraLabs \| 7.9M \| 11.95 \| 0.00 \| 28.17 \| 19.50 \| 9.22 \| 5.86 \| 11.85 \|
	\| 38 \| Tokle-3M \| techdotus \| 2.9M \| 11.89 \| 0.00 \| 24.05 \| 9.36 \| 10.32 \| 10.50 \| 18.07 \|
	\| 39 \| Supra-Mini-v3-0.5M \| SupraLabs \| 467648 \| 3.05 \| 0.00 \| 3.70 \| 0.72 \| 1.55 \| 5.88 \| 5.98 \|

	> Note: we did NOT rank every model that is ranked on the [Open SLM Leaderboard](https://huggingface.co/spaces/AxiomicLabs/Open_SLM_Leaderboard). If you have new ranks, you can post them in your model card.

	## Intended use

	Use the index to compare small base models on grammar, lexicon, facts, reading, elementary reasoning, and local coherence. Do not use it as a chat, instruction-following, or safety score. DO NOT TRAIN ON THE TASKS!!!

	## Limitations

	Language: English only. Several tasks have only a few units, so standard errors are sometimes wide. World-knowledge items reflect widely taught facts, not a neutral sample of the world. The bits-per-byte anchors are provisional. A model whose context window is shorter than an item is left-truncated, which can hurt the reading tasks but we tried to keep the tasks inside a 512 token window.

	## Credits

	Supra-μBench was built by SupraLabs. Version 0.2.0.