Google’s Gemini 3.8 Flash is constructed to “work more durable”

Google’s earlier smartest AI mannequin, Gemini 3.6 Flash, was launched in late July. The tech big wasted no time iterating on it, with Gemini 3.7 Flash launching simply weeks after. Now, simply three weeks later, the corporate is already introducing 3.7 Flash’s successor.

Google simply introduced Gemini 3.8 Flash, and this time round, the main target is on delivering next-generation intelligence for agentic workflows and cybersecurity with two totally different variants.

The brand new fashions mark Google’s third Flash launch in six weeks, with new frontier fashions nowhere in sight. The brand new fashions are Gemini 3.8 Flash and Gemini 3.8 Flash Cyber. The previous is the corporate’s “most clever workhorse mannequin,” whereas the latter is its “most succesful cybersecurity mannequin.”

Gemini 3.8 Flash is primarily geared toward builders constructing agentic functions and workflows, although additionally it is obtainable to common Google AI Professional and Extremely subscribers by way of the Gemini app.

In line with Google’s numbers, Gemini 3.8 Flash outperforms most bigger frontier fashions in relation to the DeepSWE v1.1 benchmark. The benchmark measures the flexibility to autonomously clear up complicated engineering issues, and Gemini 3.8 Flash hit 73.7% on it. That’s just under Claude Opus 5‘s 74% rating, and effectively above Gemini 3.7 Flash, Claude Sonnet 5, GPT-5.6 Sol and GPT-5.6 Terra.

Gemini 3.8 Flash additionally outperformed the entire aforementioned fashions on a number of different benchmarks, together with monetary evaluation on Vals Finance Agent v2, complicated authorized workflows on Harvey’s Authorized Agent Benchmark, data synthesis on CharXiv Reasoning, agentic terminal coding on Terminal-bench 2.1, lengthy video understanding on LVBench, and extra.

Google defined that “3.8 Flash works more durable,” and therefore the efficiency beneficial properties. On complicated duties, it reportedly displays higher diligence, with the mannequin executing further reasoning steps each time crucial. That additionally means the mannequin may, at occasions, use extra tokens to maximise efficiency.

Gemini 3.8 Flash launches at $0.75 per million enter tokens and $3.75 per million output tokens. That is introductory pricing, which can be revised to $1.50 and $7.50, respectively. Regardless, the mannequin ought to nonetheless be cheaper than counterparts like Claude Opus 5 and GPT-5.6 Sol.

Gemini 3.8 Flash Cyber, then again, isn’t obtainable to all. Entry to it’s restricted by a brand new Fairwind Program, which is able to let trusted authorities authorities, in addition to vital infrastructure operators and software program maintainers leverage the brand new mannequin.

On the industry-standard benchmark for locating vulnerabilities in C/C++, the mannequin scored 86.2%, beating its predecessor Gemini 3.5 Flash Cyber and bigger fashions like Mythos 5 and GPT-5.5-Cyber. Equally, in real-world vulnerability discovery spanning over 20 programming languages, the mannequin reached successful charge of 71%, which is considerably greater than Gemini 3.7 Flash’s 58.9% and Gemini 3.5 Flash Cyber’s 46.6%.

Elsewhere, what makes the mannequin particular is its automated patching skill. Gemini 3.8 Flash Cyber equips defenders with the flexibility to generate patches to deal with vulnerabilities. In line with the corporate’s numbers, the mannequin produced 2.6 occasions extra right patches to vulnerabilities in Chrome “than one of the best business fashions which are a lot bigger.”

For what it’s value, most common customers won’t ever actually work together with the brand new mannequin, however the enhancements it gives to defenders and builders can have an oblique affect on making your on a regular basis tech safer and higher.

Gemini 3.8 Flash is rolling out now for builders, Enterprise clients, and Google AI Professional and Extremely subscribers. Entry to Gemini 3.8 Flash Cyber, as talked about earlier, is proscribed to Fairwind Program members.