Granite Switch 4.1 30B

Use it for natural-language responses, safety-risk detection, decision classification, and multilingual speech transcription from one locally served model.

Availability

Revision 4.1.1; release date October 27, 2026

Pricing model

Not established

Deployment

Backends: HuggingFace Transformers, vLLM

Privacy

Not established

Licence

Apache 2.0

• The checkpoint includes a decision adapter with updated weights in revision 4.1.1. ^[1]

  - **Updated `decision` weights.** The `decision` aLoRA ships updated weights; its prompt format, invocation and readout are unchanged, so existing integrations work as-is.

• The embedded speech encoder transcribes English, French, German, Spanish, Portuguese, Italian, and Japanese. ^[1]

  - **Multilingual speech input.** The embedded [Granite Speech 5.0 TurboCTC](https://huggingface.co/ibm-granite/granite-speech-5.0-470m-turboctc) encoder now transcribes English, French, German, Spanish, Portuguese, Italian and Japanese, and outputs punctuated, capitalized transcripts.

• The Guardian Core adapter detects safety risks including harm, jailbreaking, profanity, violence, sexual content, social bias, and unethical behavior. ^[1]

  | [**Guardian Core aLoRA**](https://huggingface.co/ibm-granite/granitelib-guardian-r1.0/tree/main/guardian-core) | Detects safety risks: harm, jailbreaking, profanity, violence, sexual content, social bias, unethical behavior |

• The model is available as the Hugging Face checkpoint ibm-granite/granite-switch-4.1-30b. ^[1]
• The model supports HuggingFace Transformers and vLLM backends. ^[1]
• The checkpoint is licensed under Apache 2.0. ^[1]

• DeepSeek opens a real-time web search API Weekly brief 2026-W42
• One model endpoint for chat, speech, decisions, and safety checks Daily brief 2026-10-16