DAREBench - Multimodal - Multi-Step: leaderboard
Metric: Accuracy (%). Source: arxiv.org. 27 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 88.5 |
| 2 | Claude Opus 4.6 | 76.2 |
| 3 | GPT-5.5 | 75 |
| 4 | GPT-5.4 | 71.2 |
| 5 | Claude Opus 4.8 | 69.2 |
| 6 | Qwen 3.5 Plus | 69.2 |
| 7 | MiMo-V2.5 | 66.7 |
| 8 | Qwen 3.6 Plus | 61.5 |
| 9 | Claude Sonnet 4.6 | 59.6 |
| 10 | Qwen 3.6 35B A3B | 57.7 |
| 11 | Gemma 4 26B A4B (IT) | 55.8 |
| 12 | Qwen 3.6 27B | 55.8 |
| 13 | Kimi K2.5 | 54.4 |
| 14 | Qwen 3.5 27B | 53.8 |
| 15 | Gemini 3.5 Flash | 50 |
Interactive version: theaggregate.ai/benchmark?slug=darebench-multimodal-multi-step · How It Works · Data refreshed daily, snapshot 2026-09-19.