Granite Speech 5.0

IBM's Apache 2.0 speech models: a 2B ASR/translation model at 5.06% English WER and a 470M Turbo CTC model at 19,870 RTFx.

IBM · Open-weight/community models · model

Published in Feed

Overview

IBM published Granite Speech 5.0 on October 21, 2026: Granite Speech 5.0 2B, Granite Speech 5.0 Turbo CTC (about 470M parameters), and an updated Granite Speech 4.1 2B Plus, all under Apache 2.0. The 5.0 2B model transcribes English, French, German, Spanish, Portuguese, and Japanese, translates to and from English in those languages plus English to Italian and Mandarin, and averages 5.06% WER on the Open ASR Leaderboard's English short-form sets, ranking #2 among top open-weight ASR models by IBM's account. Turbo CTC reaches 19,870 RTFx, about 1.6x faster than at launch at the same accuracy.

This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.

Key features

  • 5.0 2B: 5.06% average English short-form WER on the Open ASR Leaderboard; keyword lists up to 2,000 terms; language-ID prompting
  • Translation to and from English across six languages, plus English to Italian and Mandarin
  • Turbo CTC (~470M): 19,870 RTFx, native punctuation and capitalization, seven languages
  • 4.1 2B Plus: long-form WER 13.53 to 10.22 and biased-keyword WER 11.79 to 6.48; drop-in replacement for 4.1 2B
  • Native Transformers support; the 5.0 2B also runs on vLLM, llama.cpp, and mlx-audio on Apple Silicon

Use cases

  • Self-hosted transcription and speaker-attributed meeting notes
  • Multilingual call-center transcription and translation
  • High-throughput batch ASR using the CTC model

Pricing

Free open weights under Apache 2.0.

Pricing reflects the source record and may change. Confirm current terms with the provider.

Dates and sources

Released
21 Oct 2026
Record updated
23 Oct 2026
Record last checked
23 Oct 2026

Explore the Feed

Other entries filed under open-weight/community models; inclusion does not imply equivalent features.

Browse the Feed