MCGA - Speech Reasoning: leaderboard
Multiple-choice questions about recited classical Chinese literature that need knowledge beyond the recording, answered from the audio (MCGA; Du et al., Harbin Institute of Technology, 2026), 1,000 test recordings. Accuracy in percent; higher is better.
Metric: Accuracy (%; multiple-choice questions on 1,000 test recordings that need external knowledge). Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 83.7 | #159 |
| 2 | Qwen3 Omni 30B A3B Instruct | 82.9 | #902 |
| 3 | Qwen2.5-Omni-7B | 79.3 | #1797 |
| 4 | Voxtral-Small-24B-2507 | 71.9 | #1291 |
| 5 | Voxtral-Mini-3B-2507 | 62.8 | #3702 |
| 6 | Phi-4 Multimodal Instruct | 54.4 | #3914 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mcga-speech-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-12.