Google releases Gemini 3.8 Flash for long-horizon coding

Google announced Gemini 3.8 Flash on 2 September 2026. The company describes it as its most intelligent Flash model, built for long-horizon software engineering, autonomous agents and complex enterprise workflows.

Google says this is the third Gemini Flash release in six weeks. That is the company's release cadence claim, not evidence that any other provider or model maker was responding to it. The same announcement also introduced Gemini 3.8 Flash Cyber, a separate version for trusted defenders through the Fairwind Program. This article covers the generally available Gemini 3.8 Flash model.

The Gemini API documentation lists a 1,048,576-token input limit and a 65,536-token output limit. Thinking is supported at low, medium and high effort. Minimal effort is not supported and returns an error.

The documented capabilities include caching, code execution, computer use in preview, file search, function calling, Google Maps grounding, image understanding, search grounding, structured outputs and URL context. Audio generation, image generation and the Live API are not supported by this model.

Google lists a free tier for Gemini 3.8 Flash. Paid standard pricing is $0.75 per million input tokens and $3.75 per million output tokens, including thinking tokens, through 31 December 2026. From 1 January 2027, the listed prices rise to $1.50 and $7.50 respectively.

The model code for direct API use is gemini-3.8-flash. OpenRouter lists google/gemini-3.8-flash and google/gemini-3.8-flash:batch, while OpenCode Zen lists gemini-3.8-flash. The provider registries confirm route availability, but they do not turn those routes into additional Google releases.

AIMI first observed models/gemini-3.8-flash on Google's official models endpoint at 16:46:56 SAST. Google's announcement was published at 15:00 UTC, or 17:00:00 SAST, so the endpoint was visible 13 minutes and 4 seconds before the public announcement. The first-seen time is an observation, not the release time.

AIMI first saw the standard OpenRouter route at 17:18:58 SAST, exactly 18 minutes and 58 seconds after the announcement. The batch route appeared at 17:35:00 SAST, 16 minutes and 2 seconds after the standard route. OpenCode Zen followed at 18:55:08 SAST, 1 hour, 55 minutes and 8 seconds after the announcement. The route sequence shows distribution across providers during the day, not a chain of separate model launches.

There were no earlier verified releases in this input batch to combine with Gemini 3.8 Flash. The official Google announcement and API changelog establish the maker release. AIMI's endpoint records establish when the routes became visible to the monitoring system.

Gemini 3.8 Flash is aimed at workloads where a task can run for many steps and use tools along the way. Google says the model may use more reasoning tokens at higher effort levels, so the token budget and price should be part of any production test.

Before switching a live application, test the exact provider route rather than assuming that the direct Google API, OpenRouter and OpenCode Zen behave identically. Check tool schemas, effort controls, rate limits, retention terms and current pricing. AZ Labs can provide one integration layer for teams that need to compare routes without rewriting the application each time.