IBM releases modular Granite Switch 4.1 models in 8B and 30B sizes

IBM announced Granite Switch 4.1, two modular language models at 8B and 30B parameters that combine a base model, a decision aLoRA adapter, the guardian-core safety adapter, and an embedded Granite Speech 5.0 TurboCTC encoder into a single hosted endpoint. Both checkpoints ship under Apache 2.0 with a 128K context window and support Transformers and vLLM.

• Granite-Switch-4.1-8B and 30B built on granite-4.1 bases with 128K context
• Embedded 470M-parameter Granite Speech 5.0 TurboCTC ASR encoder
• guardian-core adapter averages 1.13x better than Jev on safety tasks
• Apache 2.0 license, requires Python 3.9+ and PyTorch 2.11+