AudioProcessBench - Best-of-N Selection: leaderboard

Metric: Final-answer accuracy (%) when the critic's summed step scores pick one chain among several sampled for the same audio question; 3,872 reasoning traces (23,497 steps, 41% erroneous) from six audio and omni models on MMAR, MMSU and MMAU-Pro questions, steps labelled by two LLM annotators with human review; few-shot critic prompt with three annotated example chains, temperature 0.7; random 47.5; higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 11 models tracked.

Top models

#ModelScore
1Gemini 3 Flash59.4
2Phi-4 Multimodal Instruct52.9
3Gemma 4 E2B52.5
4Gemma 4 E4B51.6
5Qwen2.5-Omni-7B51.1
6Gemma 3n E4B50.8
7Gemma 3n E2B50.3

Interactive version: theaggregate.ai/benchmark?slug=audioprocessbench-best-of-n-selection · How It Works · Data refreshed daily, snapshot 2026-09-29.