IBM has released Granite Speech 5.0 2B, a new open speech-language model for transcription and translation — and it’s already claiming a #2 spot among open-weight models on the Open ASR Leaderboard.
Granite Speech 5.0 2B (Credits: IBM / Hugging Face)
What the new model does
The 2-billion-parameter model transcribes speech to text in English, French, German, Spanish, Portuguese and Japanese, and translates between English and those languages — plus English-to-Italian and English-to-Mandarin. It ships under the permissive Apache 2.0 license, so developers can use it commercially for free.
Under the hood, it’s a dense 2B-parameter model built on IBM’s granite-4.0-1b-base, trained on 390,000 hours of public audio plus synthetic data made for Japanese recognition, keyword biasing and translation. On the Open ASR Leaderboard, IBM reports an average English word error rate of 5.68%, with a standout 0.99% on LibriSpeech Clean.
Three models, three jobs
The 2B accuracy model isn’t alone — the release refreshes IBM’s whole speech lineup. The revised Turbo CTC model targets raw speed and on-device transcription, while the updated 4.1 2B Plus handles multi-speaker recordings with improved speaker attribution and timestamps.
All three come with native support in the Transformers library, and they run on GPUs via Transformers or vLLM, Apple Silicon via mlx-audio, or local CPU via llama.cpp. For enterprise teams that need transcription with data control and local deployment, it’s one of the strongest open options to land this year.
Also Read: find more AI model releases in our AI News section
Source: MarkTechPost — read the full story there.
