GB/T-Bench - Direct Review DMTR@10: leaderboard
Metric: Documents with at least 10 correct diagnoses (%). Source: arxiv.org. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Fable 5 | 26.09 |
| 2 | GPT-5.5 | 24.39 |
| 3 | GPT-5.6 Sol | 21.74 |
| 4 | GPT-5.4 | 11.89 |
| 5 | Qwen 3.7 Plus | 9.43 |
| 6 | Gemini 3.5 Flash | 8.2 |
| 7 | Claude Opus 4.8 | 7.17 |
| 8 | DeepSeek V4 Pro | 4.92 |
| 9 | Qwen 3 235B A22B 2507 Instruct | 4.71 |
| 10 | DeepSeek V4 Flash | 3.07 |
| 11 | GPT-5.3 | 2.66 |
| 12 | GPT-5.4 Mini | 1.43 |
| 13 | MiniMax-M2.7 | 1.02 |
| 14 | Qwen 2.5 VL 72B Instruct | 0.82 |
Interactive version: theaggregate.ai/benchmark?slug=gb-t-bench-direct-review-dmtr-10 · How It Works · Data refreshed daily, snapshot 2026-09-19.