MMBench-Live: leaderboard
MMBench-Live (2026) rebuilds MMBench with freshly collected images and automatically generated, executable-verified multiple-choice questions so the benchmark can be updated without contamination. Overall accuracy in percent, higher is better.
Metric: Overall accuracy (%; multiple-choice questions over all six L-2 abilities; 5.9K newly generated evaluation instances aligned with the MMBench task hierarchy; temperature 0). Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 7B Instruct | 83.1 | #1833 |
| 2 | Qwen 3 VL 8B Instruct | 80.41 | #981 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mmbench-live · How It Works · Data refreshed daily, snapshot 2026-11-13.