Hacking with multimodal Gemma 4 in AI Studio

Gemma 4 is now available via the Gemini API and Google AI Studio under Apache 2.0 license. It comes in two variants: a 31B dense model with 256K context window and a 26B Mixture-of-Experts model activating only ~4B parameters per inference. Both support native multimodal inputs (images + text) and chain-of-thought reasoning. Google AI Studio lets you prototype visually and export working code in TypeScript, Python, Go, or cURL with one click, including automatic base64 image encoding and thinkingConfig parameters.