HELM Audio - Air Bench Chat: leaderboard
The "Air Bench Chat" subtask of HELM Audio. Stanford HELM's audio project (AHELM): audio-language models are given speech and sound clips and scored on transcription, spoken-question answering, reasoning over audio, speaker and emotion attributes, safety and fairness.
Metric: GPT-4o Judge Critique. Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.0 Pro (Preview 02-05) | 3.25 | #305 |
| 2 | Qwen2-Audio-7B-Instruct | 3.22 | #888 |
| 3 | Gemini 2.0 Flash | 3.2 | #344 |
| 4 | Gemini 2.0 Flash (Preview) | 3.19 | #366 |
| 5 | Gemini 1.5 Pro (002) | 3.07 | #350 |
| 6 | Gemini 1.5 Flash (002) | 3.04 | #585 |
| 7 | Gemini 1.5 Flash (001) | 2.77 | #580 |
| 8 | Gemini 1.5 Pro (001) | 2.67 | #446 |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-audio-air-bench-chat · How It Works · Data refreshed daily, snapshot 2026-10-23.