DAREBench - Text - Multi-Step: leaderboard
Metric: Accuracy (%). Source: arxiv.org. 35 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 83.6 |
| 2 | Claude Opus 4.6 | 83.4 |
| 3 | MiMo-V2.5 | 82.5 |
| 4 | GPT-5.4 | 82.4 |
| 5 | MiMo-V2.5-Pro | 81.7 |
| 6 | DeepSeek V4 Pro | 80.9 |
| 7 | Gemini 3.5 Flash | 80.9 |
| 8 | Gemini 3.1 Pro (Preview) | 80.7 |
| 9 | GPT-5.4 Mini | 79.1 |
| 10 | Claude Opus 4.8 | 78.1 |
| 11 | MiniMax-M2.7 | 77.8 |
| 12 | Qwen 3.6 Plus | 77.7 |
| 13 | Qwen 3.5 Plus | 77.3 |
| 14 | Claude Sonnet 4.6 | 76.8 |
| 15 | Claude Haiku 4.5 | 76.7 |
Interactive version: theaggregate.ai/benchmark?slug=darebench-text-multi-step · How It Works · Data refreshed daily, snapshot 2026-09-19.