Xiaomi MiMo drops dialect-savvy ASR

Xiaomi is making a serious play for open ASR, not just another Whisper clone with a new coat of paint. The interesting part is the target mix: dialect coverage, code-switching, and knowledge-heavy transcription are exactly where real-world ASR still breaks.

• –The differentiator is regional Chinese speech support: Wu, Cantonese, Hokkien, and Sichuanese are much more useful for deployment in China than another English-first benchmark flex.
• –Xiaomi’s own benchmarks show MiMo-V2.5-ASR beating or matching Qwen3-ASR, FunASR, Seed-ASR, and Whisper-large-v3 across several Chinese, dialect, lyrics, and mixed-language tasks.
• –The model also claims leading average WER on the Open ASR Leaderboard’s tougher English sets, which helps it read as more than a China-only release.
• –Community reaction is split between excitement over the MIT license and skepticism about efficiency, with early Reddit comments calling out the 8B size and large weight footprint for an ASR model.
• –For developers, the practical signal is clear: Xiaomi is open-sourcing specialized speech models now, not just general-purpose LLMs, which makes the MiMo stack more credible as a full audio-text platform.