Granite Switch 4.1 combines chat, decisions, and speech in one endpoint

IBM's Granite Switch 4.1 models, Granite-Switch-4.1-8B and -30B, add a general Decision aLoRA, a Guardian aLoRA, and Granite-Speech-5.0-Turbo-CTC, which the source calls the fastest ASR model on the OpenASR Leaderboard.

All three run in one model that uses about the same GPU VRAM as the base model alone, and the adapters share its KV cache. The source says Granite-Switch-4.1-30B finished a five-step RAG workflow faster than Jev-27B.