Gemini 3.6 Flash (high)

Released on July 21, 2026, Gemini 3.6 Flash is the primary high-efficiency model in Google's Gemini 3 series. Positioned as a versatile "workhorse," it is designed to balance speed, cost, and intelligence for production-scale AI agents. This model succeeds Gemini 3.5 Flash, introducing a knowledge cutoff of March 2026 and achieving a reported 17% reduction in output token usage through more efficient reasoning steps and tool-call orchestration.

Thinking Effort Tiers

The "High" variant of Gemini 3.6 Flash refers to the High Thinking Effort tier, one of three customizable reasoning levels (Low, Medium, and High) available to developers. While the base model is optimized for low-latency tasks, the High Effort mode enables the model to allocate more compute to complex problems, such as full-stack code refactoring, spatial reasoning, and multi-step agentic execution. This configuration allows the model to compete with larger frontier models in specific benchmarks while maintaining a Flash-class pricing structure.

Capabilities and Performance

Gemini 3.6 Flash is natively multimodal, supporting inputs across text, images, video, audio, and PDF documents. It features a 1-million-token context window with a 64,000-token output limit. In technical evaluations, the model has demonstrated significant improvements in coding precision and computer-use tasks, scoring 83.0% on OSWorld-Verified and 49% on the DeepSWE benchmark. It is specifically optimized to produce production-ready code with fewer unwanted edits and execution loops compared to previous generations.

Developer Features

The model supports advanced agentic workflows through features like native function calling, a built-in computer-use tool, and specialized performance in visual chart reasoning. Google also introduced specialized variants alongside the core 3.6 Flash release, including Gemini 3.5 Flash-Lite for high-throughput messaging and Gemini 3.5 Flash Cyber for security-focused code analysis.