# Google launched Gemini 3.5 Flash for agentic and coding workflows > The first Gemini 3.5 model became available across consumer, developer and enterprise surfaces with emphasis on long-running agent tasks. Canonical URL: https://www.devobs.io/news/news-gemini-3-5-flash-launch/ By: Sofia Reyes Published: 2026-09-06T11:58:54.627Z Updated: 2026-09-06T11:58:54.627Z Event date: 2026-05-19 Section: AI Google introduced Gemini 3.5 on May 19, beginning the family with Gemini 3.5 Flash. The [launch post](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/) positioned Flash for coding and long-horizon agentic work and said it was available in the Gemini app, AI Mode in Search, Google Antigravity, the Gemini API in Google AI Studio, Android Studio and enterprise products. Google also said Gemini 3.5 Pro was still in internal use and planned for a later rollout. ## Model upgrades need task-level evaluation For engineering teams, broad availability is less important than whether a model changes the failure profile of an existing workflow. An agent that runs longer can accumulate more tool calls, state and irreversible effects. Better benchmark scores do not answer whether it retries safely, respects application permissions or recovers after a partial tool failure. Treat 3.5 Flash as a new runtime dependency. Pin the model identifier where the API permits it, save representative traces and compare success, latency, token use, tool selection and human correction against the version it may replace. Include long tasks that cross context boundaries rather than testing only isolated prompts. ## Roll out agents behind bounded authority A production trial should separate reasoning quality from action safety. Give the model read-only tools first, cap steps and wall-clock time, and require application-side checks for every mutation. Test duplicate calls, stale tool results, interrupted sessions and adversarial content returned by a tool. Google's release spans consumer and developer products, but availability does not make their controls identical. Builders should verify the API's current model documentation and terms for their surface, then promote gradually with a rollback path to the prior model and retained traces for regression analysis. ## Source references -