Atlas

Models

← All models

Qwen3 Omni 30B A3B Instruct

Alibaba · 2025-09-22 · 35.3B parameters

Qwen3-Omni-30B-A3B-Instruct is Alibaba's non-thinking Qwen3-Omni checkpoint; it contains the MoE Thinker and Talker together with audio and vision encoders. It accepts text, image, audio, and video input and produces text and real-time streaming speech, with Qwen reporting support for 119 text languages, 19 speech-input languages, and 10 speech-output languages.

Benchmark scores

BenchmarkScore
AA-LCR0.0
AA-Omniscience Index-69.2
AIME 202552.3
Artificial Analysis Intelligence Index5.1
Artificial Analysis Omniscience Accuracy14.4
Artificial Analysis Omniscience Hallucination Rate97.7
Artificial Analysis Openness Index44.4
Audio MultiChallenge (Audio Output)24.3
Capability130.8
CritPt0.0
GPQA Diamond62.0
Humanity's Last Exam (Text-Only)5.1
IFBench (Artificial Analysis)31.2
LiveCodeBench (Artificial Analysis)42.2
MMLU Pro72.5
MMMU Pro55.5
SciCode (Artificial Analysis)18.6
Terminal-Bench Hard1.5
τ²-bench (Telecom)16.4
Loading Atlas data…