Skip to content

IBM updates Granite Switch: new ASR model, Decision adapter

Short answer

IBM released Granite Switch 4.1.1, a patch to the 4.1 line. It carries an updated multilingual speech-to-text model covering 10 languages and an improved Decision adapter. It is fully backwards compatible with 4.1, and because the base model is untouched, teams can swap in the new adapters without migrating.

What this means for operators

If you run sales calls or a support line, the speech model matters most: transcripts arrive with punctuation and capitalization, so you can pass them straight into summarization or routing steps without a cleanup pass. Because 4.1.1 is backwards compatible and leaves the base model alone, a team already on 4.1 can swap in the new adapters and re-run its own call samples, with no migration project. The source gives no pricing, licence or accuracy figures for the new speech model, so test it on your own recordings and languages before relying on it. The Decision adapter gain is reported only against IBM's own test set, so treat it as unverified for your workflows until you have checked it.

IBM's Granite team has published Granite Switch 4.1.1 on the Hugging Face blog, just over a week after Granite Switch 4.1. The company calls it a patch release and uses the version number the way software teams do: it is fully backwards compatible with 4.1, needs no migration and brings incremental improvements.

What changed

Two components were updated:

  • Granite Speech ASR model. It converts speech to text in 10 languages and produces native punctuation and capitalization. IBM says transcripts are therefore ready for downstream models without a separate formatting step.
  • Decision adapter in granitelib-core. IBM reports better overall performance on the Jev Decision Index test set. That is IBM's own benchmark result as stated in the post, and the post does not give figures.

Why the release process is the story

The post stresses how quickly the update came together. Only the new ASR model and the Decision adapter had to be trained. Nothing else was recomputed, and the base model was not touched. Users compose the adapter versions they want and deploy them.

This follows from the Switch design introduced with 4.1, in which a language model is configured with specialized adapters that keep the cache efficient, adding new expertise on a single endpoint. A fix to one capability therefore does not require replacing the whole model.

What IBM says comes next

IBM is working on tutorials and an Adapter SDK. Developers and outside vendors would be able to create their own adapters, roll out new versions, trial features, and release updates in the same manner they use for ordinary software. No release date is given for either, and the post only says more will follow.

Where to try it

The post points to three models and one demo Space. The demo, Speech to Decisions with Granite, shows live speech being heard and sized up by one Granite Switch model, and it showcases the new multilingual and punctuation capabilities.

What is not in the source

The post does not list the 10 languages, publish accuracy results for the speech model, or describe licence terms or pricing. Any decision to adopt the update should rest on testing with your own audio.

Source: Hugging Face · In the Atlas: Hugging Face →

Next step

Visibility Analyzer

An AI news item will not tell you how assistants see your own site. The Visibility Analyzer checks that on your live site: which answers cite you, which pages an engine cannot retrieve, and what to fix first. Free to run.

Run a free visibility audit

Free to run. No card.

Fee
Free
Length
One run, minutes

Free tier: two analyses a day, no card required.