M-GATE - Grammatical Error Detection: leaderboard
Binary grammatical-error detection on linguist-written, adversarially selected sentences in 30 languages, from M-GATE (arXiv 2608.03803v1, Appendix Table J.1, frozen July 2026 snapshot). Score is the mean Matthews correlation coefficient across languages at temperature 0 (-1 to 1, 0 = chance; higher is better); each reasoning configuration is a separate row.
Metric: Mean MCC over 30 languages (-1 to 1; temperature 0). Source: arxiv.org. Saturation forecast: Around 2029. 82 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 0.36 | #64 |
| 2 | Claude Fable 5 | 0.36 | #20 |
| 3 | Gemini 3.5 Flash | 0.3 | #66 |
| 4 | Gemini 3.5 Flash (High) | 0.28 | #66 (Gemini 3.5 Flash) |
| 5 | Gemini 3 Flash (Preview) | 0.23 | #90 |
| 6 | Grok 4.20 | 0.2 | #179 |
| 7 | Gemini 3.5 Flash (Minimal) | 0.19 | #66 (Gemini 3.5 Flash) |
| 8 | GPT-5.5 (High) | 0.18 | #33 (GPT-5.5) |
| 9 | Gemini 3 Flash (Preview) (Minimal) | 0.18 | #90 (Gemini 3 Flash (Preview)) |
| 10 | GPT-5.5 (Medium) | 0.16 | #33 (GPT-5.5) |
| 11 | Claude Opus 4.6 (Thinking) | 0.15 | #67 (Claude Opus 4.6) |
| 12 | Gemini 2.5 Pro | 0.11 | #159 |
| 13 | Claude Sonnet 5 | 0.1 | #65 |
| 14 | Claude Opus 4.8 | 0.09 | #39 |
| 15 | Gemini 2.5 Flash | 0.08 | #255 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=m-gate-grammatical-error-detection · How It Works · Data refreshed daily, snapshot 2026-11-05.