Datasets:
Link to model predictions (ActiveVision-Results)
Browse files
README.md
CHANGED
|
@@ -97,6 +97,12 @@ axis, not toward the human band. Tool-using coding agents reach at most 50.6%
|
|
| 97 |
(Claude Code, Fable 5) at $2.74–$7.63 and 12–15 minutes per item — a human answers
|
| 98 |
unaided in 34 seconds.
|
| 99 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 100 |
## Citation
|
| 101 |
|
| 102 |
```bibtex
|
|
|
|
| 97 |
(Claude Code, Fable 5) at $2.74–$7.63 and 12–15 minutes per item — a human answers
|
| 98 |
unaided in 34 seconds.
|
| 99 |
|
| 100 |
+
## Model predictions
|
| 101 |
+
|
| 102 |
+
Every model's final answer on every item, with per-task scores for all leaderboard
|
| 103 |
+
configurations, is in [activevisionai/ActiveVision-Results](https://huggingface.co/datasets/activevisionai/ActiveVision-Results).
|
| 104 |
+
Browse them on the [per-task results page](https://activevision.dev/results.html).
|
| 105 |
+
|
| 106 |
## Citation
|
| 107 |
|
| 108 |
```bibtex
|