Gemini 3.8 Flash (high)

Gemini 3.8 Flash is a multimodal large language model developed by Google DeepMind, released on September 2, 2026. Positioned as a "workhorse" model for production environments, it was designed to balance high-speed inference with significant gains in complex reasoning and autonomous agent performance. It follows a rapid release cycle, arriving just three weeks after Gemini 3.7 Flash, and serves as the primary intelligence layer for Google's agentic workflows and coding platforms.

The model's architecture is optimized for long-horizon tasks, particularly in software engineering and multi-step tool use. According to technical reports, Gemini 3.8 Flash delivers substantial performance improvements on benchmarks such as DeepSWE v1.1 and OSWorld-2.0, outperforming several larger frontier models in autonomous engineering tasks. While it shares the same foundational core as its predecessor, it is refined through recursive agentic evaluation loops that allow it to execute extra reasoning steps and call tools iteratively for higher accuracy.

Gemini 3.8 Flash features a context window of approximately 1 million tokens and supports multimodal inputs including text, image, audio, and video. It maintains three distinct "thinking levels" (Low, Medium, and High) that allow developers to adjust the model's reasoning effort based on the complexity of the request. A specialized variant, Gemini 3.8 Flash Cyber, was released concurrently for vulnerability detection and automated patching, though it is restricted to authorized defenders through the Fairwind Program.