BASIS-Bench - ASR Fair-Speech Accent: leaderboard
Table 8 content tasks evaluate spoken-question answering and speech transcription across demographic and accent slices. These are generated content scores; the separate demographic classification and stereotyping metrics are not imported here. ASR uses the published 100 minus WER, whose lower bound is unbounded.
Metric: 100-WER. Source: arxiv.org. Saturation forecast: Around December 2026. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3 Omni 30B A3B Instruct | 95.3 | #364 |
| 2 | Voxtral-Small-24B-2507 | 92.3 | #569 |
| 3 | Qwen2.5-Omni-7B | 83.9 | #608 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=basis-bench-asr-fair-speech-accent · How It Works · Data refreshed daily, snapshot 2026-10-16.