IBM releases Granite Switch 4.1.1 with updated speech-to-text and decision adapters
Patch release adds multilingual speech recognition with native punctuation and improved decision-making performance, while maintaining full backward compatibility with Granite Switch 4.1.
1 source · cross-referenced
- Granite Switch 4.1.1 is a patch release that updates the Granite Speech ASR model and the Decision adapter without retraining the base model.
- The updated Granite Speech ASR model supports multilingual speech-to-text across 10 languages with native punctuation and capitalization.
- The Decision adapter in granitelib-core shows improved performance on the Jev Decision Index test set.
- The update is fully backward compatible with Granite Switch 4.1 and requires no migration.
- IBM plans to release tutorials and an Adapter SDK to enable third-party development of adapters.
IBM released Granite Switch 4.1.1, a patch release that updates components of its modular Granite Switch architecture without altering the base model. The update includes an updated Granite Speech ASR model and an improved Decision adapter in granitelib-core.
The Granite Speech ASR model now supports multilingual speech-to-text across 10 languages and includes native punctuation and capitalization, producing transcripts ready for downstream models without additional formatting steps.
The Decision adapter in granitelib-core has been updated to improve overall performance on the Jev Decision Index test set, according to the announcement.
Granite Switch 4.1.1 is fully backward compatible with Granite Switch 4.1 and requires no migration, aligning with software versioning conventions. IBM noted that no recomputation was necessary beyond training the new ASR model and Decision adapter.
IBM indicated that this modular update mechanism can extend beyond patch releases, enabling developers to compose and deploy the latest adapters without retraining the base model. The company is working on tutorials and an Adapter SDK to allow third-party providers and developers to build and iterate on their own adapters.
The announcement highlighted the efficiency of the update process, emphasizing that the base model remained untouched and that new features could be composed and deployed rapidly.
- Nov 11, 2026 · Hugging Face
Hugging Face proposes token pooling-aware training to improve compressed multi-vector retrieval
Trust79 - Nov 11, 2026 · TechCrunch — AI
Fractile raises $720M Series C to develop AI chips and compiler stack
Trust79 - Nov 11, 2026 · The Verge — AI
Google’s Tabletop AI experiment lets users build custom collaborative apps via waitlist
Trust79