Back to Model
G

Granite Switch 4.1

Model Oct 15, 2026

IBM's modular 8B and 30B checkpoints combining Granite chat, Decision and Guardian activated LoRA adapters, and an English speech-to-text encoder in one serving endpoint.

Visit primary link 3 sources attached
IBM's conceptual Granite Switch illustration with a large base-model block and three smaller adapter blocks.
IBM's conceptual adapter illustration from its English announcement. The blocks explain modular composition; they are not measured memory sizes or benchmark results. Source: IBM Research / IBM Granite
decision-modelsself-hostedmultimodal

What happened lately

Granite Switch 4.1 timeline

The Granite Switch 4.1 speech-and-decision configuration has a model-card release date of October 14, 2026, followed by IBM’s October 15 announcement. Both sizes combine a Granite 4.1 base, Decision and Guardian Core adapters, and an embedded 470M English transcription encoder. Earlier preview variants have a different adapter bundle without this audio configuration.

Applications select an adapter through the chat template; IBM attributes efficiency to reusing base-model context caches. Performance and memory comparisons remain developer claims. Speech is transcribed to text before entering the language model, is English-only and is not jointly trained with it. The included Guardian adapter does not supply the entire library’s factuality and policy-checking capabilities. Custom adapter training through a planned Mellea SDK is not yet a shipped feature of these checkpoints.