FMSU-Bench: leaderboard
Metric: Average score (%) over the 14 tasks in both the Chinese and English subsets (13 multiple-choice accuracies and the PATA transcription score), over the 20,000-plus bilingual FMSU-Bench items; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 74 |
| 2 | Gemini 3 Flash | 71.9 |
| 3 | Gemini 2.5 Flash | 69 |
| 4 | Qwen2.5-Omni-7B | 59.7 |
Interactive version: theaggregate.ai/benchmark?slug=fmsu-bench · How It Works · Data refreshed daily, snapshot 2026-10-07.