Google launches Gemini 3.8 Flash AI model on September 2 with one million token context window

Google released its Gemini 3.8 Flash artificial intelligence model on September 2. The company positioned this version as a general-availability tool for software engineers building autonomous agents and managing complex enterprise workflows. Google stated the model combines improved reasoning and coding capabilities with lower latency costs found in previous Flash iterations. The new system accepts inputs including text, images, video, audio, and PDF files while producing only text output. A defining feature is a context window supporting approximately one million input tokens and up to 65,536 output tokens. Google confirmed the model supports multiple functions such as code execution, file search, function calling, structured outputs, search grounding, URL context, and computer use. While some features remain in preview mode, others are ready for production use through developer and enterprise platforms. Internal evaluations show Gemini 3.8 Flash outperformed most larger frontier models on the DeepSWE v1.1 benchmark regarding long-horizon software engineering tasks. The company highlighted stronger performance compared to the previous Gemini 3.7 Flash version in handling extended tasks involving multiple files and repeated tool calls.