Mistral Large 4 Preview - SciCode-Verified pass@1 (n=6): leaderboard
Mistral Large 4 public-preview post (2026-10-06): SciCode-Verified scientific-coding problems as run by Mistral, pass@1 estimated from six samples per problem, for eleven models. Main-problem or subproblem scoring and the background setting are not stated, and Mistral's comparator values differ from the SciCode-Verified repository's own results (GLM-5.2 88.1 here, 84.3 there), so this board is kept apart from them. pass@1 (%), higher is better.
Metric: pass@1 (%, n=6). Source: mistral.ai. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 94.2 | #6 |
| 2 | Qwen3.8 2.4T A95B | 93.8 | #59 |
| 3 | GLM-5.3 | 92.5 | #44 |
| 4 | MiMo-V2.6-Pro | 91.9 | #48 |
| 5 | Mistral Large 4 | 91.8 | #106 |
| 6 | Claude Opus 5 | 91.3 | #16 |
| 7 | DeepSeek V4 Pro (0813) | 91 | #58 |
| 8 | Kimi K3 | 90.3 | #35 |
| 9 | GLM-5.2 | 88.1 | #80 |
| 10 | Mistral Medium 3.5 | 70 | #375 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=mistral-large-4-preview-scicode-verified-pass-1-n-6 · How It Works · Data refreshed daily, snapshot 2026-11-05.