Google made Gemini 3.8 Flash available on September 2, three weeks after the previous version. The model supports 1,048,576 input tokens and 65,536 output tokens, with a focus on coding, long-running software tasks, and AI agents. The same API prices as Gemini 3.7 Flash will apply through the end of 2026; developers can move to production with the stable gemini-3.8-flash identifier. Google released Gemini 3.6 Flash on July 21 and 3.7 Flash on August 13, delivering three versions in six weeks. The model processes text, images, video, audio, and PDF files and generates text; it offers various development tools, Search and Maps grounding, URL context, and structured output. The reasoning level can be adjusted; computer use is in preview. It scored 61.4% on Vals Finance Agent v2 and 54.9% on HLE-Verified; it surpassed many major models on DeepSWE v1.1.
Why it matters
The back-to-back releases are bringing to the forefront the question of which model developers using Gemini should prefer in production and how transitions should be managed. Maintaining the same API prices as the previous version through the end of 2026 and providing a stable model identifier make it easier for teams seeking to update their applications to assess the risks of cost and version changes. Long-context support, multiple input types, and structured output features could affect the design of systems that carry out coding and long-running software tasks in particular. However, the fact that the computer use feature is still in preview leaves open which capabilities should be considered production-ready in agent-based use cases. Although the test results provide a basis for comparing performance, their scope regarding behavior under different real-world usage conditions remains limited.
Background
Google is not a new name in the FikirPilot archive: we have published 20 articles mentioning it in the past 90 days; the latest was dated September 5, 2026.