Google released Gemini 3.1 Ultra with a 2-million-token context window—double the 1M limit in Gemini 3.0 Ultra. The model adds native sandboxed Python execution: it writes code, runs it, reads output, and revises within a single inference call. Multimodal reasoning across charts, diagrams, and video frames is improved over the prior generation. Gemini 3.1 Ultra is available through Gemini Advanced, Google AI Studio, and the Gemini API.