AudioProcessBench - First-Error Identification: leaderboard

Metric: PRMScore (0-100) for locating the first erroneous step of each reasoning chain; 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; higher is better. Source: arxiv.org. Saturation forecast: Around October 2027. 11 models tracked.

Top models

#ModelScore
1Gemini 3 Flash63.5
2Gemma 4 E4B35.5
3Gemma 3n E4B25.7
4Gemma 4 E2B24.7
5Qwen2.5-Omni-7B19.3
6Gemma 3n E2B18.8
7Phi-4 Multimodal Instruct18.5

Interactive version: theaggregate.ai/benchmark?slug=audioprocessbench-first-error-identification · How It Works · Data refreshed daily, snapshot 2026-09-29.