0.5B ASR model hits 6.96% multilingual WER, beating Meta's 15x-larger omniASR
thisguyknowsai · x · 2026-11-19
Per an official Hugging Face post, a tiny 0.5B-parameter speech recognition model is punching far above its weight:
- 40+ languages with automatic language detection; runs in Transformers and vLLM with no custom code
- 6.96% average error rate on the Open ASR Leaderboard multilingual track, ahead of Meta's omniASR-LLM 7B (15x larger) and Whisper large-v3
- 6.6% WER on the English track
- 7.86% WER on a 302-hour call-centre benchmark spanning 14 English accents, essentially matching NVIDIA's 30B Nemotron 3.5 Transcribe at 7.78%
Small models are rapidly closing in on — and beating — much larger ones on ASR.
More from Models
- Anthropic allegedly profiles users and silently downgrades them to weaker models at full price — JafarNajafov · 2026-11-19
- Antigravity CLI 1.5.12 screenshot shows Gemini 4 Neon (Medium) model option — patloeber · 2026-11-19
- Subscription value test: Gemini ~26x vs OpenAI ~10.5x API credits; real-bug eval underway — PawelHuryn · 2026-11-19
- Daily digest: Xiaomi MiMo spinoff rumor, Tencent cuts Yuanbao spend, Gemini 4 Neon ships — 快鲤鱼 · 2026-11-19
- Branching a ChatGPT workspace wipes the agent's context, user complains — LucaAmb · 2026-11-19
- tinygrad publishes 10-411 week-1 assignment, testing LLMs one-shot on held-out test set — CatAstro_Piyush · 2026-11-19