List your company

Audio Question Answering on Multi-sound-object soundscapes

72.1AccuracyQwen2.5-Omni
33.15243.263553.37563.4865Jan 14, 2026

Evaluation results

MethodLinks
Qwen2.5-Omnisetting=GRPO with external audio data2026.0172.175.1
Speech-Handsmode=parallel agentic, sampling=majority sampling2026.0159.477.37
Speech-Handsmode=parallel agentic, setting=SFT with official training data2026.0158.2975.75
AudioFlamingo 3 (AF3)2026.0157.3174.49
Qwen2.5-Omni2026.0156.3257.87
Qwen2.5-Omnisetting=GER: => AF3 (cascaded agentic)2026.0152.0268.93
Gemini-2-Flash2026.0146.3456.61
Qwen2.5-Omnisetting=GRPO with official training data2026.0139.4365.54
Qwen2.5-Omnisetting=SFT with official training data2026.0134.6563.13