Google Gemini

Gemini 3.1 is Google DeepMind's March 2026 flagship model update, arriving on March 20th. It represents the most significant architectural advance in the Gemini series: unlike previous versions that treated multiple modalities as add-ons to a text-primary model, Gemini 3.1 was built from pretraining to reason natively across text, image, audio, and video within a single context window.

The headline capability is a fully utilized 2-million token context window that works across all modalities — you can feed it hours of video, audio transcripts, images, and text documents simultaneously, and it reasons coherently across all of them.

Key features: - 2M token context window, fully functional across all modalities - Native multimodal reasoning: no transcription intermediaries for audio/video - 77.1% on ARC-AGI-2 (doubles its predecessor's reasoning score) - 94.3% GPQA Diamond — expert-level scientific knowledge - Built-in Code Execution tool: runs and tests code in a sandboxed environment mid-conversation - Improved grounding for significantly reduced hallucinations on factual queries - Available via Google AI Studio and Vertex AI at unchanged /2 per million tokens (input/output) - Gemini 3.1 Ultra available on Google AI Ultra plan

Gemini 3.1 is especially strong for enterprise use cases involving large document analysis, multimodal workflows (combining images, audio, video, and text), and complex reasoning chains. The Code Execution sandbox makes it a powerful choice for data analysis, scientific workflows, and education.