Announced September 2, 2026, Gemini 3.8 Flash is a harder-working coding and agents model Google says tops larger rivals on some tasks, shipping alongside a defense-focused 3.8 Flash Cyber — at the same fast, cheap price as 3.7 Flash.
2026-09-02 · by Moe Ameen
Published 2026-09-02 · Updated
Short answer
Gemini 3.8 Flash is a fast, low-cost Google model for coding, agents and analysis, announced on September 2, 2026. Google says it "works harder" by taking extra reasoning steps and calling tools again and again. It costs $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, then $1.50 and $7.50. A sibling, 3.8 Flash Cyber, goes only to vetted security teams.
Key takeaways
On September 2, 2026, Google announced Gemini 3.8 Flash, describing it as a workhorse model built to "work harder" — putting in extra reasoning steps and more iterative tool use to push through complex, multi-step engineering and analysis tasks. It is the third release on Google's cost-efficient Flash tier in roughly six weeks, following Gemini 3.6 Flash in late July and Gemini 3.7 Flash on August 13. Like those, it is a text-and-reasoning model for coding, agents, and knowledge work — not an image or video generator, despite the "Flash" name it shares with Google's separate media models.
Google reports gains over 3.7 Flash and says the model outperforms several larger frontier models on some benchmarks while staying cheap. The company highlights autonomous engineering (a test it labels DeepSWE v1.1) plus stronger results in quantitative and professional domains, citing the Vals Finance Agent V2 and Harvey's Legal Agent Benchmark.
Google also shipped a specialized sibling, Gemini 3.8 Flash Cyber, tuned for defensive security — vulnerability discovery and automated patch generation — which Google reports reaches frontier-level results (for example around 47.2% pass@1 on a patching benchmark it calls CWE-Bench and a real-world vulnerability-discovery success rate above 70% across 20 programming languages). Access to the Cyber variant is limited to trusted defenders through Google's Fairwind Program. As always with launch-day figures, treat vendor benchmarks as Google's own until third parties test them.
Pricing matches 3.7 Flash's introductory offer: $0.75 per million input tokens and $3.75 per million output through December 31, 2026, then $1.50 per million input and $7.50 per million output starting January 1, 2027. Gemini 3.8 Flash is available in the Gemini app for Google AI Pro and Ultra subscribers, in AI Mode, in Gemini for Google Sheets, and to developers through the Gemini API, Google AI Studio, Android Studio, and Google Antigravity, plus Gemini Enterprise.
The pitch on 3.8 Flash is that it is more diligent — it takes extra reasoning steps and chains more tool calls to finish a hard task. That is genuinely useful for the front of a content workflow: hand it a stack of sources and it will summarize, extract, and plan a week of angles more carefully than a lighter model. But diligence is not distribution. A more thorough plan is still just a plan — the model records no video, designs no carousel, writes no caption in your voice, and posts to nothing. That gap is structural to a language model, and it did not narrow with this release.
Kompozy is the layer that closes it. Take the researched angles or rough scripts you generate with 3.8 Flash and drop them in: Kompozy rewrites them in your voice through a Persona Brief, then generates the finished formats a model can't — captioned Persona Shorts and HeyGen avatar video, brand-exact carousels and quote cards via HyperFrames, photo posts, blog articles, and email newsletters — and schedules and publishes the whole set across the eight social platforms plus blog and email on Autopilot. Pair the cheap, careful brain with the engine that turns its plan into a published content week.
This persona short came straight out of the Kompozy workspace. It is here to show the format, not this news item.
Best for:
Not ideal for:
Google announced it on September 2, 2026. It is a fast, low-cost model for coding, agents and knowledge work that Google says "works harder" by taking extra reasoning steps and calling tools iteratively. It handles text and reasoning, not images or video.
It is a sibling model tuned for defensive security work such as finding vulnerabilities and patching them. Google reports 47.2% pass@1 on CWE-Bench and says access is limited to trusted defenders through its Fairwind Program.
It costs $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. From January 1, 2027 the price rises to $1.50 and $7.50.
Google lists the Gemini app for AI Pro and Ultra subscribers, AI Mode in Search, Gemini in Google Sheets, the Gemini API, Google AI Studio, Android Studio, Google Antigravity and Gemini Enterprise.
It can draft and plan text, but it produces no images, video or scheduled posts. Turning its drafts into finished posts across platforms is a separate job that a content engine like Kompozy handles.
We checked every date, figure and product claim on this page against the sources listed below on 2026-10-23. Model details and prices come from Google's announcement; the 3.6 Flash and 3.7 Flash dates come from the launch coverage listed. Kompozy plan and format details come from kompozy.io/pricing.
Sources