List your company

Audio Understanding on MMAU (Speech, Sound, Music, Overall)

80.4Overall ScoreQwen3.5-Omni
28.29641.82355.3568.877Apr 14, 2026Apr 21, 2026Apr 29, 2026May 6, 2026May 14, 2026May 21, 2026May 29, 2026

Evaluation results

MethodLinks
Qwen3.5-OmniOpen-Source=false, Size=Flash, Reasoning mode=Reasoning off2026.0480.4————
Qwen3-OmniOpen-Source=true, Size=30B-A3B, Reasoning mode=Reasoning off2026.0477.5————
Nemotron 3 Nano OmniOpen-Source=true, Size=30B-A3B, Reasoning mode=Reasoning off2026.0476.972.8—74.2—
Step-Audio2Audio Input Architecture=Continuous2026.0472.768.279.368.4—
Qwen2.5-OmniAudio Input Architecture=Continuous2026.0471.570.678.165.9—
UAS-AudioAudio Input Architecture=Continuous2026.0469.4677071.3—
Kimi-AudioAudio Input Architecture=Continuous2026.0468.262.275.766.8—
UniAudio-TokenBackbone=Qwen2.5-3B, Tuning strategy=tuned identically2026.0561.145.0570.2767.96—
UAS-Audio-DAudio Input Architecture=Discrete2026.045646.2559.1662.57—
GLM-4-Voice-TokenizerBackbone=Qwen2.5-3B, Tuning strategy=tuned identically2026.0555.243.2460.0662.28—
CosyVoice2Backbone=Qwen2.5-3B, Tuning strategy=tuned identically2026.0554.739.9461.5662.57—
StableTokenBackbone=Qwen2.5-3B, Tuning strategy=tuned identically2026.0553.245.0558.5655.99—
WavTokenizerBackbone=Qwen2.5-3B, Tuning strategy=tuned identically2026.0551.736.9460.3657.78—
GLM-4-VoiceAudio Input Architecture=Discrete2026.0430.335.4427.6327.84—