Artificial intelligence news.
Straight from the original announcements.

IBM Granite Switch 4.1 Aims to Serve Chat, Decisions and Speech From One Endpoint

October 15, 2026, 22:34Source: Hugging Face
IBM Granite Switch 4.1 Aims to Serve Chat, Decisions and Speech From One Endpoint
Image: Hugging Face

IBM's Granite team has published a Hugging Face blog post titled "Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint". The title points to a single endpoint covering conversation, decision-making and speech tasks, though the text available for this report contains few technical details.

Key points
  • IBM's Granite team published "Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint" on Hugging Face.
  • The post links to five models, two Spaces and one collection.
  • The models include a 0.5B speech recognition model and text generation models listed at 30B and 9B.
  • The collection holds four adapters that extend Granite models.
  • The available page text gives no benchmarks, pricing or availability details.

What the post covers

The post appears on the ibm-granite organisation's Hugging Face blog. Its title describes three capabilities served from one endpoint: chat, decisions and speech. The accessible page text does not include the body of the article, so how Granite Switch routes between these functions, and what "decisions" means in practice, cannot be confirmed from the material available.

The page does show that the article links to five models, two Spaces and one collection. That is the clearest sign of what accompanies the release.

Models and adapters linked in the article

The listing includes an automatic speech recognition model at 0.5B parameters, which fits the speech part of the title. Alongside it are text generation models, including entries marked 30B and 9B. The page text does not give names, benchmark results or licensing for any of them.

The single collection is described as a set of adapters that extend Granite models. It contains four items. Adapters are typically used to add specialised behaviour to a base model, but the source does not explain how Granite Switch uses them, so that link should not be assumed.

Related Granite work

The page also lists two earlier posts from the same author, both dated August 25, 2026. One is "Granite 4.2 LLMs: How They're Built". The other is "Extremely Fast and Accurate Transcription with Granite Speech 5.0 Turbo CTC". They suggest IBM has been developing language and speech models in parallel, and Granite Switch 4.1 brings both areas under one title.

What it means

On the evidence available, Granite Switch 4.1 is IBM's effort to present text and speech capabilities as a single service rather than separate model deployments. Details on performance, pricing, availability and supported languages are not in the source text. Readers who want to evaluate the release should check the full post and the linked model pages on Hugging Face.

Related news

IBM Granite Switch 4.1.1 Pitches Model Updates Like Software Updates
ModelsOctober 29, 2026, 14:37

IBM Granite Switch 4.1.1 Pitches Model Updates Like Software Updates

IBM's Granite team has published Granite Switch 4.1.1 on Hugging Face, a point release presented under the idea of updating models the way developers update software.…

IBM Granite Speech 5.0 targets faster, more accurate speech recognition
ModelsOctober 21, 2026, 16:58

IBM Granite Speech 5.0 targets faster, more accurate speech recognition

IBM's Granite team has published Granite Speech 5.0 on Hugging Face, a post titled "Transcribe, Translate, and Attribute more Accurately and Faster". The page lists…

PyTorch 2.15 brings FP8 and much faster MoE decoding to Macs
For developersOctober 20, 2026

PyTorch 2.15 brings FP8 and much faster MoE decoding to Macs

PyTorch 2.15 adds float8_e4m3fn support on the MPS backend and, according to the Hugging Face PyTorch team, makes mixture-of-experts models decode 5 to 6.5 times faster…