Omni-DeepSearch - Speech: leaderboard

Omni-DeepSearch (2026) tests audio-driven deep search: from one or more audio clips alone, a model must infer the audio entity, call text, image and video search tools and give a short verifiable answer, judged by a majority of three LLM judges. This board is accuracy on the 120 single-audio questions whose clip is speech, in percent; higher is better.

Metric: Accuracy (%) on the 120 single-audio questions whose clip is speech, audio-driven deep search: starting from audio clips alone, the model calls text, image and video search tools (at most 10 text or image searches and 3 video searches per question) and gives a short verifiable answer, judged by majority vote of GPT-5.4, Gemini-3-pro and Claude-Sonnet-4.6; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Pro55#145
2Gemini 2.5 Pro25#276
3Gemini 3 Flash20#175
4MiMo-V2.515.83#383
5Qwen 3.5 Omni Plus14.17#254
6MiMo-V2-Omni10.83#465
7Qwen 3 Omni 30B A3B (Thinking)5.83
8Gemini 2.5 Flash Lite4.17#964
9Qwen3.5 Omni Flash4.17#602
10Qwen2.5-Omni-7B1.67#1884
11Qwen3 Omni 30B A3B Instruct1.67#930
12Qwen2.5-Omni-3B0#3519

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=omni-deepsearch-speech · How It Works · Data refreshed daily, snapshot 2026-11-21.