Gemini 3.8 Flash is out, Google's third Flash release in six weeks

Google’s Flash line now ships faster than most companies ship changelogs.

What launched

Google released Gemini 3.8 Flash on September 2, 2026, describing it as “our most intelligent workhorse model, delivering significant improvements from 3.7 Flash across software engineering, agentic tasks, and critical, multi-step reasoning in specialized domains.” By Google’s own count it is the third Flash version in six weeks. The model scores 54.9% on HLE-Verified and, Google says, outperforms most larger frontier models on the DeepSWE v1.1 benchmark for autonomously solving engineering problems.

It is available through Google AI Studio, the Gemini API, the Gemini app, AI Mode in Search, Google Sheets and Google Antigravity.

The price, and its expiry date

Pricing holds at $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. From January 1, 2027 the standard rate applies: $1.50 and $7.50, exactly double. Anyone building on the introductory price has a four-month window before their bill changes.

A second variant, Gemini 3.8 Flash Cyber, is described as Google’s most capable cybersecurity model and goes to trusted government authorities, critical infrastructure operators and software maintainers through the Fairwind Program rather than to the public.

The pattern

Three Flash models in six weeks is a cadence, not an event. Google is using its cheapest tier to keep pace with rivals’ frontier releases while holding the price line until the end of the year. The question for developers is whether the January price step lands before or after the next Flash.

How it edges Opus 5

On the comparison tables Google published, 3.8 Flash comes in ahead of Claude Opus 5 on HLE-Verified (54.9% vs 54.4%), the Vals Finance Agent v2 benchmark (61.4% vs 58.6%) and Harvey’s legal agent benchmark (10.0% vs 6.7%). Google credits the gains to what it calls diligence: on complex tasks the model executes extra reasoning steps and calls tools repeatedly, and it may use more tokens at higher effort levels. For efficiency-first workloads Google says developers can lower the effort setting or stay on 3.7 Flash, which remains supported. Reviewers who read the model card note that 3.8 Flash is built on 3.7 Flash rather than a new base, so the score gain and the token cost arrive together, ahead of the price doubling on January 1, 2027.