OmniClean - UNO-Bench: leaderboard
The 228 of 1,000 UNO-Bench queries (multiple-choice omni-modal questions (the UNOBench-MC subset)) that OmniClean (StepFun-Audio Team, 2026) keeps after removing those Qwen3-VL-30B-A3B-Thinking can answer from the visual input and question alone. Accuracy in percent under UNO-Bench's official protocol; higher is better.
Metric: Accuracy (%; the 228 of 1,000 UNO-Bench multiple-choice (UNOBench-MC) queries left after visual-only cleaning; official answer matching). Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3 Omni 30B A3B Instruct | 29.17 | #902 |
| 2 | Qwen2.5-Omni-7B | 24.84 | #1797 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=omniclean-uno-bench · How It Works · Data refreshed daily, snapshot 2026-11-12.