The announcement was published on October 21, 2026. All three models are released under Apache 2.0.
The new 2B-parameter model succeeds Granite Speech 4.1 2B. IBM says it transcribes English, French, German, Spanish, Portuguese, and Japanese; it translates to and from English for those languages, with English-to-Italian and English-to-Mandarin translation also listed. It supports prompt-based language identification, optional punctuation and capitalization, and keyword lists of up to 2,000 terms. IBM says its training data includes 390,000 hours of public audio plus synthetic data.
The 470M-parameter Turbo CTC update adds transcription in seven languages, native punctuation and capitalization, and claimed throughput of 19,870 RTFx, up from roughly 12,600 at launch. The updated 4.1 2B Plus retains its architecture, prompt format, and API; IBM reports a 24% reduction in long-form WER, a 45% reduction in biased-keyword errors with 20-term lists, and average timestamp shift falling from 210.86 to 98.54.
The source reports IBM's own benchmark and improvement figures. It does not establish independent performance validation, deployment cost, hardware requirements, or comparative results outside the cited tests.