Google Rolls Out Gemini 3.8 Flash, Third Flash Update in Three Months

Google began rolling out Gemini 3.8 Flash on September 2, 2026, just three weeks after its previous Flash release — the third update to the Flash line in three months. The model is already live for Google AI subscribers in the Gemini app, as well as in Google Antigravity and AI Studio.

Google describes 3.8 Flash as its "most intelligent Flash model," engineered specifically for long-horizon software engineering tasks, autonomous agents, and complex enterprise workflows. According to Google, the model beats its predecessor across "various benchmarks," though the company has not published specific scores alongside the announcement.

Pricing and availability

Google is offering an introductory rate of $0.75 per 1 million input tokens and $3.75 per 1 million output tokens, matching the pricing structure used for the prior Flash release. That introductory pricing holds through December 31, 2026, after which standard rates are expected to apply — though Google has not disclosed what those will be.

Knowledge cutoff stays split

As with the previous Flash model, Gemini 3.8 Flash carries an inconsistent training cutoff. Google states the model's knowledge extends to March 2026 "for some domains," while users may find the model's knowledge limited to January 2025 in others. This split-cutoff disclosure has now appeared in back-to-back Flash releases, suggesting Google is continuing to blend fresher fine-tuning data with an older base training set rather than fully retraining the model from scratch each cycle.

No changes to context window size have been disclosed with this release, and Google has not published parameter counts, as is standard practice for the Gemini line.

What this means

Three Flash-tier releases in three months signals Google is treating Flash as its highest-velocity iteration lane — a place to ship incremental gains in coding and agentic performance without the fanfare of a Pro-tier launch. That cadence is unusual even by current industry standards, where most labs space major or minor model updates weeks to months apart, and it raises a real question: how much of this is genuine model improvement versus rapid A/B testing of variants against benchmarks and pricing tiers.

The split knowledge cutoff — March 2026 for some domains, January 2025 for others — is the detail worth watching. It suggests Google is not doing full retrains between releases, instead patching or fine-tuning a shared base model to hit agentic and coding benchmarks faster. For developers building on Flash, that means checking task-specific outputs for recency gaps rather than assuming a uniform cutoff.

The held-over introductory pricing through year-end keeps Flash competitive on cost against Anthropic's and OpenAI's comparable fast-tier models, but the real signal is velocity: if Google sustains a three-week release cycle, Flash becomes less a stable product and more a continuously tuned service, which changes how enterprises should think about version pinning and regression testing in production.