Granite Switch 4.1.1: Updating Models Like Software
Granite Switch 4.1.1: Updating Models Like Software Published October 29, 2026 Granite Switch 4.1 introduced a modular Switch architecture that allows language models to be conf...
By Software Development Team
Granite Switch 4.1.1: Updating Models Like Software
Published October 29, 2026
Granite Switch 4.1 introduced a modular Switch architecture that allows language models to be configured with specialized, cache-efficient adapters. These adapters add new capabilities while keeping them available through a single endpoint.
Granite Switch 4.1.1 is a patch release for Granite Switch 4.1. As with a software patch release, version 4.1.1 is fully backwards compatible with version 4.1 and does not require migration. The update provides incremental improvements without changing the base model.
Changes in version 4.1.1
The release includes two updates:
- Updated Granite Speech ASR model: Granite Speech 5.0 470M TurboCTC provides multilingual speech-to-text for 10 languages. It also adds native punctuation and capitalization, allowing transcripts to be passed to downstream models without a separate formatting step.
- Updated Decision adapter: The Decision adapter in granitelib-core delivers improved overall performance on the Jev Decision Index test set.
Adapter-based updates
The update demonstrates how the Switch architecture can reduce the work required to release model improvements. Other than training the new ASR model and Decision adapter, no components were recomputed. The base model remains unchanged, while users can compose the adapters they need and deploy the resulting configuration.
This approach is intended to support more than patch releases for IBM-provided models. IBM is developing tutorials and an Adapter SDK that will allow developers and third-party providers to create their own adapters, iterate on versions, test features, and release updates using a software-like workflow.
The latest models are available in the Granite Switch collection. The multilingual speech recognition and punctuation features are also used in the Speech-to-Decisions demo.