SCOPED - CountBenchQA: leaderboard
SCOPED v1 Table1; final4B/8B independently distilled models from exactQwen3VLInstruct backbones, no skill text at student inference. Exact eval release xiongyq/SCOPED-Eval revision0b43c3d2b5cd3238b4c735976ebaefc28db6d6b9; counts491,527,1790,2500,1168,1740,2600,1000; PixMo540annotations minus13missingimages is published527cohort. VisualPuzzles source splittrain;WeMath testmini;CCOCR recognitiononly. Overall arithmetic mean of eight percentages. Metric: Accuracy (%). Higher is better. Source: https://arxiv.org/html/2610.22217v1
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 91.24 | #78 |
| 2 | Claude Sonnet 4.6 | 90.63 | #89 |
| 3 | Claude Opus 4.6 | 89.21 | #60 |
| 4 | GPT-5.2 | 88.39 | #109 |
| 5 | Qwen 3 VL 8B Instruct | 83.91 | #414 |
| 6 | Qwen 3 VL 4B Instruct | 81.47 | #510 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=scoped-countbenchqa · How It Works · Data refreshed daily, snapshot 2026-10-17.