Models · Embodied interaction

Can it understand people and what is happening around it?

Omni-modal models that see, hear and talk: they follow speech, understand sound and video together and know when things happen, so a robot can work alongside people.

Daily-Omni

Average accuracy on audio-visual questions, %

#1 · Daily-OmniAgiBot WITA-Omni 1.086.22%

1AgiBot WITA-Omni 1.086.22%

2AgiBot WITA-Omni Preview85.21%

3Qwen3.8-Omni-Flash (thinking)85.1%

4Qwen3.5-Omni-Plus84.68%

5Gemini 3.1 Pro Preview82.79%

6Doubao Seed 2.0 Lite82.12%

7NVIDIA Nemotron 3 Nano Omni 30B74.52%

8Qwen3-Omni-30B-A3B-Thinking73.6%

Source: Daily-Omni project page · snapshot 2026-10-30 · Daily-Omni authors (Zhou, Wang, Wu, Jiang) · Results as published by each leaderboard; we have not rerun them

MMAU

Average accuracy on the test split, % (MMAU-v05.15.25)

#1 · MMAUAudio-Thinker75.98%

1Audio-Thinker75.98%

2Nova 2 Omni75.28%

3Step-Audio-273.86%

4MiMo-Audio72.59%

5Audio Flamingo 372.42%

6Qwen2.5-Omni71%

7Step-Audio-2-mini70.23%

8Gemini 2.5 Pro69.36%

Source: MMAU leaderboard · snapshot 2026-10-30 · University of Maryland (MMAU authors) · Results as published by each leaderboard; we have not rerun them

TimeLens-Bench

Average across three refined grounding sets (recall at IoU 0.3/0.5/0.7 and mIoU)

#1 · TimeLens-BenchAgiBot WITA-Omni 1.066

1AgiBot WITA-Omni 1.066

2TimeLens2-8B65

3TimeLens2-4B65

4Gemini 2.5 Pro63

5TimeLens2-2B61

6TimeLens-8B60

7Gemini 2.5 Flash57

8Qwen3-VL-235B-A22B57

Source: TimeLens-Bench leaderboard · snapshot 2026-10-30 · Tencent ARC Lab · Results as published by each leaderboard; we have not rerun them

Want your model on our bench, or a benchmark we don't cover yet? We run custom evaluations.Talk to us →