IBM Granite Switch 4.1.1 patch updates speech ASR and Decision adapter, base model untouched
IBM released Granite Switch 4.1.1, a patch release that is fully backwards compatible with 4.1 and requires no migration. It updates the Granite Speech ASR model (10 languages) and the Decision adapter in granitelib-core, while the base model stays unchanged.
IBM has released Granite Switch 4.1.1, a patch update that is fully backwards compatible with Granite Switch 4.1 and requires no migration. The release updates two components, the Granite Speech ASR model and the Decision adapter in granitelib-core, and leaves the base model untouched. IBM announced it on Hugging Face on October 29, 2026.
What changed in 4.1.1
- Updated Granite Speech ASR model: multilingual speech-to-text across 10 languages, with native punctuation and capitalization. IBM says transcripts are ready for downstream models without a formatting pass.
- Updated Decision adapter in granitelib-core: IBM says it delivers improved overall performance on the Jev Decision Index test set. The post gives no score for the improvement.
IBM describes the version number as carrying its usual software meaning: backwards compatible with 4.1, with incremental improvements under the hood.
How the update was built
According to IBM, the notable part of the release is how quickly and efficiently the teams shipped it. Apart from training the new ASR model and the Decision adapter, nothing was recomputed. The base model is unchanged. Users compose the latest adapters they want and deploy.
This follows Granite Switch 4.1, which IBM released just over a week earlier. That release introduced the modular Switch architecture, which lets users configure a language model with specialized, cache-efficient adapters, adding new expertise on a single endpoint. IBM's Hugging Face posts also list Granite Speech 5.0, published October 21, 2026, and Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint, published October 15, 2026.
What is coming next
IBM says it is working on tutorials and an Adapter SDK so that developers and third-party providers can build their own adapters. The stated aim is to let them iterate on versions, test features, and ship updates the way software is shipped. IBM gave no release date, saying only "More soon."
IBM also points to a Speech-to-Decisions demo that uses the new multilingual and punctuation capabilities. The post links three models, one Space and one collection on Hugging Face.
Pricing, context window, parameter counts and benchmark scores are not stated in the post.
What this means
The 4.1.1 release shows the Switch architecture being used as IBM intends: individual adapters and the ASR model are retrained and swapped while the base model stays fixed. IBM says this meant nothing else had to be recomputed. For teams already running Granite Switch 4.1, the practical effect is that the update is described as a drop-in change, with no migration needed.
The planned Adapter SDK is the part to watch. If it ships as described, third-party providers would be able to publish adapter versions on the same pattern, rather than only IBM-provided ones. Until the SDK and tutorials appear, that remains a stated plan.
Related Articles
IBM Granite Switch 4.1 8B and 30B fold chat, decisions and speech-to-text into one endpoint
IBM has released Granite-Switch-4.1-8B and Granite-Switch-4.1-30B, models that bundle a Decision adapter, a Guardian safety adapter and a speech-to-text model behind a single endpoint. IBM says the combined model uses roughly the same GPU VRAM as the base model alone. Benchmark claims are IBM's own, and context window and pricing were not disclosed.
IBM's Granite Timeseries Ensemble pools four open models and edges each member on GIFT-Eval point-forecast MASE
IBM Research released Granite Timeseries Ensemble R1, a recipe that combines four pre-trained Granite forecasters into one median and quantile forecast with no retraining or fitted weights. On 97 GIFT-Eval configurations, the Probabilistic Ensemble scores 0.683 MASE, against 0.685 for the best single member.
IBM Granite Speech 5.0 2B ranks #2 among open ASR models at 5.06% WER under Apache 2.0
IBM released granite-speech-5.0-2b, a 2B-parameter speech-language model trained on 390,000 hours of public audio. IBM claims it ranks #2 among open-weight ASR models with a 5.06% average English WER. IBM also updated its 470M Turbo CTC model to 19,870 RTFx and refreshed the weights of granite-speech-4.1-2b-plus.
IBM and Kaggle Partner on Continuously Updated ScarfBench Leaderboard for Java Migration Agents
IBM and Kaggle are partnering on a continuously updated public leaderboard for ScarfBench, IBM's open benchmark for coding agents doing enterprise Java framework migrations. The benchmark covers migrations across Jakarta EE, Quarkus, and Spring, according to an IBM Research technical note published October 27, 2026.
Comments
Loading...