Granite Speech 5.0 2B

IBM's open-weight speech model for multilingual transcription and translation.

Also known as granite-speech-5.0-2b, Granite Speech

What it does

Granite Speech 5.0 2B is IBM's speech-language model for multilingual transcription and translation, released October 21, 2026. IBM names it granite-speech-5.0-2b and says it has the same parameter count as granite-speech-4.1-2b. It transcribes English, French, German, Spanish, Portuguese, and Japanese. The announcement says it translates to and from English in those languages, plus English to Italian and Mandarin. It was trained on 390,000 hours of public audio. New controls include language-ID prompting and keyword lists of up to 2,000 terms. The weights are Apache 2.0, with support in Transformers, vLLM, llama.cpp, and mlx-audio.

Who is Granite Speech 5.0 2B for?

Teams who need open-weight transcription or speech translation, including on their own hardware.

Tags

• speech
• open-weight
• transcription