Gemini 3.8 Flash reaches GA for long-horizon agents

Google has made Gemini 3.8 Flash (gemini-3.8-flash) generally available in the Gemini API. The September 2, 2026 release positions the model for long-horizon software engineering, autonomous agents, and complex enterprise workflows.

A new Flash workhorse

Google describes Gemini 3.8 Flash as its most intelligent Flash model, engineered for coding and agent workflows that need to stay coherent across longer tasks. It is the latest step in the 3.x Flash line after Gemini 3.7 Flash reached general availability in August.

The model is intended for applications that need a balance of intelligence, latency, and cost rather than the largest model available for every request. Teams should compare it with the models already in their routing policy using representative tasks, tool calls, and failure recovery—not only short prompt tests.

Available in Google’s developer products

The Gemini API changelog marks the model as GA, and Google’s Antigravity announcement says Gemini 3.8 Flash is available inside Google Antigravity. Antigravity’s post lists the familiar Flash price of $0.75 per million input tokens and $3.75 per million output tokens for its product context.

Pricing and limits can vary by surface and service tier, so API users should consult the current model and pricing documentation before copying the Antigravity figures into an API budget. Keep the exact model ID in configuration and test fallback behavior if the model is unavailable in a selected region or product.

Related September API changes

Google’s Gemini API changelog also adds agentic video understanding for Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. The models can navigate long video timelines by requesting transcripts, frames, or audio tracks as needed, which Google says can use up to 88% fewer tokens than static processing in long-form cases.

That capability is separate from the Gemini 3.8 Flash model release. Applications using video should validate the actual token savings and output quality on their own media, especially where timestamps, speaker changes, or visual evidence matter.

Lyria 3.5 enters public preview

Google’s September 3 Gemini API changelog also lists Lyria 3.5 in public preview. The lyria-3.5 model supports text and image inputs and generates full-length, 44.1 kHz stereo music with controls for duration and structure.

Lyria 3.5 is a separate music-generation release, not a Gemini 3.8 Flash capability. Treat it as an API preview: test licensing, latency, output controls, and production suitability before building a user-facing music workflow around it.

Migration checklist

Update the model identifier deliberately, then rerun evaluations for coding, tool use, long-context retrieval, and multi-step agent recovery. Monitor latency, token usage, and completion quality after rollout, and keep a fallback model available while the new GA model is incorporated into production routing.

Sources

• Google AI for Developers — “Release notes | Gemini API” (September 2, 2026): https://ai.google.dev/gemini-api/docs/changelog
• Google Antigravity — “Gemini 3.8 Flash in Google Antigravity” (September 1, 2026): https://antigravity.google/blog/gemini-3-8-flash-in-google-antigravity
• Google for Developers on X — official developer updates: https://x.com/googledevs
• Google AI for Developers — “Release notes | Gemini API” (September 3, 2026): https://ai.google.dev/gemini-api/docs/changelog