M3-Bench - L4: leaderboard
Metric: Score. Source: huggingface.co. 36 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek V3.2 Speciale | 94 |
| 2 | GPT-5 Pro | 92 |
| 3 | Gemini 3 Deep Think | 90 |
| 4 | Claude Opus 4.5 | 86.8 |
| 5 | GPT-5.1 | 86.2 |
| 6 | Gemini 3 Pro | 83.5 |
| 7 | Kimi K2 (Thinking) | 81 |
| 8 | Grok 4.1 | 79 |
| 9 | GPT-OSS-120B | 77.7 |
| 10 | DeepSeek V3.2 | 76.2 |
| 11 | Mistral Large 3 | 70.5 |
Interactive version: theaggregate.ai/benchmark?slug=m3-bench-l4 · How It Works · Data refreshed daily, snapshot 2026-09-19.