GB/T-Bench - Diagnosis Recall - Scope Consistency: leaderboard
GB/T-Bench (arXiv 2608.06312v1): review of 488 Chinese national standard documents seeded with 7,306 errors across 25 error types, by direct LLM review without the authors' multi-agent reviewer. Exact diagnosis recall for errors in the scope review dimension, such as scope statements inconsistent with the content, percent (Table 3). Higher is better.
Metric: Diagnosis recall (%; exact section, dimension and error-type match). Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 81.81 | #33 |
| 2 | GPT-5.6 Sol | 75.96 | #22 |
| 3 | Claude Fable 5 | 70.87 | #20 |
| 4 | GPT-5.4 | 69.65 | #83 |
| 5 | Qwen 3.7 Plus | 67 | #98 |
| 6 | Gemini 3.5 Flash | 60.88 | #66 |
| 7 | DeepSeek V4 Pro | 60.33 | #112 |
| 8 | GPT-5.3 | 54.3 | #104 |
| 9 | DeepSeek V4 Flash | 53.38 | #155 |
| 10 | GPT-5.4 Mini | 43.33 | #228 |
| 11 | Claude Opus 4.8 | 41.96 | #39 |
| 12 | Qwen 2.5 VL 72B Instruct | 38.39 | #383 |
| 13 | Qwen 3 235B A22B 2507 Instruct | 36.11 | #315 |
| 14 | MiniMax-M2.7 | 33.46 | #270 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=gb-t-bench-diagnosis-recall-scope-consistency · How It Works · Data refreshed daily, snapshot 2026-11-05.