GB/T-Bench - Diagnosis Recall - Normative References: 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 normative-reference review dimension, which needs cross-section and cited-standard checks, percent (Table 3). Higher is better.
Metric: Diagnosis recall (%; exact section, dimension and error-type match). Source: arxiv.org. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 72B Instruct | 60.4 | #383 |
| 2 | GPT-5.4 Mini | 48.28 | #228 |
| 3 | GPT-5.5 | 43.89 | #33 |
| 4 | Qwen 3 235B A22B 2507 Instruct | 43.47 | #315 |
| 5 | DeepSeek V4 Flash | 43.16 | #155 |
| 6 | Claude Fable 5 | 42.86 | #20 |
| 7 | Qwen 3.7 Plus | 42.84 | #98 |
| 8 | GPT-5.4 | 39.6 | #83 |
| 9 | DeepSeek V4 Pro | 39.39 | #112 |
| 10 | Gemini 3.5 Flash | 38.35 | #66 |
| 11 | Claude Opus 4.8 | 36.57 | #39 |
| 12 | GPT-5.6 Sol | 34.07 | #22 |
| 13 | GPT-5.3 | 33.02 | #104 |
| 14 | MiniMax-M2.7 | 31.77 | #270 |
Interactive version: theaggregate.ai/benchmark?slug=gb-t-bench-diagnosis-recall-normative-references · How It Works · Data refreshed daily, snapshot 2026-11-05.