IBM Granite Switch 4.1: modular 8B/30B models with chat, decisions, and speech in one endpoint
IBM's modular open models (8B and 30B, Apache 2.0) composing a base model with decision, safety, and speech adapters into a single hosted endpoint.
Why it matters
One checkpoint switches between chat, typed decisions, safety screening, and transcription using cache-efficient Activated LoRAs — at roughly the VRAM cost of the base model alone.
What you could build with it
An enterprise IT team could deploy one Switch endpoint to handle employee chat, safety-flagged decisioning, and call transcription for a support desk instead of three separate models. The result: three workloads on one GPU with built-in guardrails and shared KV-cache reuse.
Does it hold up?
Too early to judge — announced Oct 15; the compose-at-VRAM-cost claims are IBM's own, with a Speech-to-Decisions HF Space demo but no independent deployment reports yet.
Built with IBM Granite Switch 4.1
- IBM Releases Modular Granite Switch 4.1 Models in 8B and 30B Sizesunite.ai · Breakdown of the 8B/30B configs, the decision and guardian-core safety aLoRAs, and the embedded Granite Speech 5.0 TurboCTC encoder.
- granite-switch-4.1-30b-preview model cardhuggingface · Full adapter composition docs: 12 adapters from the Granite Libraries collection activated on demand via control tokens.
- hammer/labs May sweep: ibm/granite-switch-4.1 filedgithub · Independent model-tracking repo flagged the Switch architecture as a novel enterprise deploy pattern — 12 LoRA experts in one checkpoint.
Learn more
First spotted on hf: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74StepFun Step-5-Preview — open BF16 weights releasedStepFun published the full BF16 weights for its flagship Step-5-Preview agentic model (600B-parameter sparse MoE, 27B…models · JEV 0.73ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72Mistral Large 4 (Le Chonk): 1.05T-param open-weight MoEMistral's new flagship: a 1.05T-parameter mixture-of-experts multimodal model with 49B active params and 1M-token…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71JetBrains Mellum 2.1: 12B MoE coding model trained for agentsJetBrains released Mellum 2.1, a 12B MoE model (2.5B active parameters, 128K context) under Apache 2.0, trained mainly…models · JEV 0.69
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.