FMSU-Bench: leaderboard
FMSU-Bench (2026) tests fine-grained speech understanding in Chinese and English with multiple-choice questions whose distractors include subtle acoustic confusions and text-only semantic traps, plus a tagged transcription task. The overall score averages the 14 tasks over both languages, 13 accuracies and one transcription score; higher is better.
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 February 2027. 10 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 74 | #99 |
| 2 | Gemini 3 Flash | 71.9 | #175 |
| 3 | Qwen 3 Omni 30B A3B | 69.4 | #847 |
| 4 | Gemini 2.5 Flash | 69 | #463 |
| 5 | MiMo-Audio | 64.1 | #931 |
| 6 | Qwen2.5-Omni-7B | 59.7 | #1884 |
| 7 | Kimi-Audio | 54.3 | #1735 |
| 8 | FMSU-Bench Step-Audio 2 (checkpoint unspecified) | 48.7 | |
| 9 | Audio Flamingo 3 | 47.6 | |
| 10 | Qwen2-Audio (FMSU-Bench checkpoint unspecified) | 23.6 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=fmsu-bench · How It Works · Data refreshed daily, snapshot 2026-11-21.