Speech Reasoning on MMAR-Speech
72.11Speech AccuracyGemini 2.5 Flash
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini 2.5 FlashSize=-2026.03 | 72.11 | |
| GPT-4o AudioSize=-2026.03 | 70.41 | |
| TGSRSize=7B2026.03 | 64.63 | |
| Audio-ThinkerSize=8.4B2026.03 | 64.29 | |
| Qwen2.5-omni + Reasoning SFTSize=7B2026.03 | 62.93 | |
| Qwen2.5-omni + Only STASize=7B2026.03 | 61.22 | |
| Qwen2.5-omniSize=7B2026.03 | 59.86 | |
| Audio Flamingo 3Size=7B2026.03 | 57.48 | |
| Audio-CoTSize=8.4B2026.03 | 34.01 | |
| Audio-ReasonerSize=8.4B2026.03 | 32.99 | |
| SALMONNSize=7B2026.03 | 24.35 |