Gemma 4 31B
by Google
Best for: Low-cost dense multimodal work with tool calls and optional reasoning.
Gemma 4 31B is Google DeepMind's 30.7B-parameter dense open-weight model. It takes text and images, calls tools natively, and has a configurable thinking mode and a 262,144-token context window. We serve it in fp8.
Best for: Low-cost dense multimodal work with tool calls and optional reasoning.
