Google Releases Gemini 3.8 Flash AI Model, Third Launch in Six Weeks With Cybersecurity Focus

Google releases Gemini 3.8 Flash, its third Flash model in six weeks

Google hasn't released a frontier-level Gemini Pro AI model since early 2026, but it sure loves rolling out new Gemini Flash variants. Today, Google is announcing its third Flash model release in just six weeks, making it more likely that we'll never see the promised Gemini 3.5 Pro. But no matter, says Google, because Gemini 3.8 Flash is its best reasoning and coding model yet. Gemini 3.8 Flash comes in two flavors. There's the standard Flash, which Google describes as a "workhorse" model that's good for anything from agentic tasks to software development. Then we have Gemini 3.8 Flash Cyber, which runs on the same foundations but has been tuned for vulnerability detection and mitigation. For developers, Google has the same pitch as it did for the 3.7 Flash release just a couple of weeks ago. API access to the model is available at an "introductory rate" through the end of the year: $0.75 per million input tokens and $3.75 per million output tokens. The regular price will be $1.50 / $7.50, but it's likely there will be new models available long before the price changes. Google probably sees the lower prices as a necessity given that other AI labs have recently dropped token pricing to keep increasingly wary businesses engaged with AI tools. Google has offered its usual raft of benchmark numbers, which appear to show Gemini 3.8 Flash competing with (or even beating) larger and more expensive models. The numbers show a marginal improvement over Gemini 3.7 Flash in most tests, but the gains are larger in coding evaluations. Gemini 3.8 Flash is now at the top of the DeepSWE leaderboard, which measures a model's ability to solve complex software engineering problems, and it does so at a lower cost (at the current discounted rate). Google reportedly delayed the release of Gemini 3.5 Pro when its coding performance couldn't match other models, but if these numbers reflect reality, even Google's new Flash models are competing with the market leaders. Computer use has been a struggle for Google's models. While Gemini 3.8 Flash is an improvement over 3.7 Flash in the OSWorld-2.0 test of agentic computer use, it's still far behind the market leader Claude Opus. In fairness, GPT isn't great in this test, either. Gemini 3.8 Flash Cyber, which replaces the 3.5 version, isn't something most of us will ever need to think about, but these models are increasingly important in niche fields. Google claims that the new cyber-security model has demonstrated a substantial improvement over its previous models with internal testing. Gemini 3.8 Flash Cyber reportedly identified more vulnerabilities and issued working patches more often. The Chrome security team apparently saw a 2.6x increase in patch accuracy with the new model. The Cloud team reports that 3.8 Flash Cyber found a critical vulnerability in just two hours. Google also has statements from partners like Wiz and Palo Alto Networks attesting to the power of Gemini 3.8 Flash Cyber. Gemini 3.8 Flash will be available across the Google ecosystem starting today, but Gemini 3.8 Flash Cyber is currently limited to trusted testers and governments. Like the past Flash release, you'll need a Pro or Ultra subscription to access Gemini 3.8 Flash in the Gemini app, but you can always visit AI Studio if you want to tinker with it for free.

Google says its new Gemini 3.8 Flash model 'works harder' but might cost more

Google launched Gemini 3.8 Flash, arriving just a few weeks after its predecessor. The company claims the new model "works harder" than Gemini 3.7 Flash by performing more reasoning steps on complex tasks and "calling tools iteratively." It has the same introductory pricing as 3.7 Flash, $0.75 per million input tokens and $3.75 per million output tokens, but could still end up costing users more. Google warns that "the model might use more tokens to maximize performance, especially at higher effort levels." Developers can keep using Gemini 3.7 Flash if they want to minimize token usage. Gemini 3.8 Flash's launch was followed by some early impressions online. Artificial Analysis highlighted the model's pricing, saying Gemini 3.8 Flash is "the cheapest we've measured at this level of intelligence. This is up ~40% from Gemini 3.7 Flash despite unchanged per-token pricing, driven by a 30% increase in output tokens per task and more turns on agentic evaluations." Aigora.ai CEO John Ennis compared Gemini 3.8 Flash to Anthropic's models, saying it offers "Opus 5 coding quality but at a fraction of the cost and super fast," adding, "This is going to be so awesome for things like making remotion videos." Google says Gemini 3.8 Flash offers "significant improvements" for software engineering and autonomous AI agents. It outperforms its predecessor and other frontier models on the DeepSWE v1.1 software engineering benchmark, including Anthropic's Fable 5, which got an upgrade earlier this week that also promises stronger performance at a lower price, by cutting the price to use cached data. Google's new model also outperformed its competitors on the Vals Finance Agent V2 benchmark and Harvey's Legal Agent benchmark. However, the new model also "ships with safeguards against misuse in the domains of Chemical, Biological, Radiological, and Nuclear (CBRN) and cyber offense." Google also released Gemini 3.8 Flash Cyber alongside its new Fairwind Program, which is limited to governments and "trusted partners." The program, whose 650 members include CrowdStrike and the Center for Internet Security, offers access to 3.8 Flash Cyber and Google's CodeMender agent, which Google says can help "autonomously find and fix vulnerabilities, protecting critical infrastructure, public services, and national security." Gemini 3.8 Flash is available now for consumers with a Google AI Pro or Ultra subscription, as well as developers and enterprise users.

With Gemini 3.8 Flash, Google reminds everyone it's still in the race

Google on Wednesday announced the release of Gemini 3.8 Flash in an attempt to reclaim its reputation as a top-tier model maker in the process. The company's commitment to the frontier model race has been in doubt since at least early August when Google DeepMind CEO Demis Hassabis stepped down to become chairman of the AI research biz and Koray Kavukcuoglu took over leadership under a less exalted title, SVP. Google CEO Sundar Pichai's reassurances weren't particularly convincing in light of the company's failure to release Gemini 3.5 Pro in June as promised. Nor were they helped by the modest benchmark metrics of Gemini 3.5 Flash, which, while speedy and cost-efficient, have been overshadowed in terms of intelligence scores by open-weight models from Chinese AI companies over the past few months. But Google has been iterating rapidly. Gemini 3.8 Flash is its fourth Flash model in as many months and it has improved enough to be considered alongside the highest scoring models in terms of intelligence while also being performant and relatively affordable - at least until its introductory price ($0.75/M input tokens, $3.75/M output tokens) doubles in the new year. "Gemini 3.8 Flash delivers substantial gains from 3.7 Flash, often approaching the performance of higher-cost frontier models," said Tulsee Doshi, Google senior director of product management, and Raluca Ada Popa, Gemini Security Lead at Google DeepMind, in a blog post. "On DeepSWE v1.1 (Long-Horizon Software Engineering) 3.8 Flash outperforms most larger frontier models in autonomously solving complex engineering problems end to end, only at a fraction of the cost." Gemini 3.8 Flash, set to high reasoning, scores 59 on the Artificial Analysis Intelligence Index, an increase of three points from its predecessor. That puts it level with GPT-5.6 Sol (extra high, 59) and Grok 4.6 (medium, 59). From there, current intelligence rankings, based on nine benchmarks, are: GLM-5.3 (max, 60), Kimi K3 (max, 60), Grok 4.6 (high, 61), GPT-5.6 Sol (max, 61), Claude Opus 5 (max, 63), and Claude Fable 5.1 (max, 66). At $0.58 per Intelligence Index task, Gemini 3.8 Flash is "the cheapest model at its level of intelligence," according to Artificial Analysis. As a point of comparison, Anthropic's latest general usage flagship model, Fable 5.1, costs about 6x more - $3.76 per Intelligence Index task. Gemini 3.8 Flash costs about 40 percent more per task than its similarly priced predecessor because it tends to output more tokens and to execute more turns when acting as an agent. "3.8 Flash works harder," explain Doshi and Popa. "On complex tasks, it exhibits greater diligence - executing extra reasoning steps, and calling tools iteratively. At times, the model might use more tokens to maximize performance, especially at higher effort levels." The two Googlers also note that Gemini 3.8 Flash shows improvements in benchmarks relevant to enterprise knowledge work, like Vals Finance Agent V2, Harvey's Legal Agent Benchmark, and HLE-Verified. Gemini 3.8 Flash is available for developers through Google Antigravity, the Gemini API in Google AI Studio and Android Studio, and interface design service Stitch. Enterprises can use Gemini Enterprise. Consumers with Google AI Pro and Ultra subscriptions can access the model via the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets. Google has also launched the Fairwind Program to provide governments, critical infrastructure organizations, and software maintainers with access to Gemini 3.8 Flash Cyber, a version of the model tuned for software vulnerability hunting and remediation. ®

Google releases Gemini 3.8 Flash and a cybersecurity variant limited to governments

Gemini 3.8 Flash arrives three weeks after 3.7 with a cybersecurity variant limited to trusted testers and unnamed governments, while the AI Act attaches documentation, copyright and training data obligations to every general purpose model placed on the EU market Google has released Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, its third Flash release in six weeks, with the cyber model restricted to trusted testers and governments. Under the EU AI Act every general purpose model placed on the market carries documentation, copyright and training data obligations, and a systemic risk model must be notified to the Commission within two weeks. Google has released Gemini 3.8 Flash. It is the company's third Flash model in six weeks, Ars Technica reported. There are two variants. A general model Google calls a workhorse, and Flash Cyber, tuned for finding and fixing software vulnerabilities. Flash Cyber replaces the 3.5 version. The Pro line has not moved since early 2026. TNW reported that the cheap model is now two versions ahead of the flagship. Pricing is introductory until the end of the year. $0.75 per million input tokens and $3.75 output, against $1.50 and $7.50 afterwards. Rivals have been cutting token prices to keep wary businesses engaged. In Europe each release is a compliance event. Article 53 requires technical documentation, a copyright policy and a public training data summary for every general purpose model placed on the market. Google signed up to this willingly. It joined the general purpose AI Code of Practice on 30 July 2025, a week after Meta refused to. The systemic risk tier carries a clock. A model trained above 10 to the 25th floating point operations must be notified to the Commission within two weeks under Article 52. That window is shorter than the gap between these releases. Gemini 3.7 Flash arrived three weeks before this one. Whether any of them crosses the threshold is not public. Google has not said, and the presumption turns on training compute rather than on benchmark results. Flash models are by design smaller than the Pro line. The cyber variant raises a different question. Flash Cyber goes only to trusted testers and governments, and Google does not say which governments. The rest is the familiar weak spot. Google's own figures put Flash behind Claude Opus on agentic computer use, a year after TNW covered the computer use tool added at 3.5. The security claims are all internal. A 2.6x improvement in patch accuracy for the Chrome team, and a critical vulnerability found in two hours, both measured by Google. Which is what the cadence obscures. Every release carries its own European obligations, for a product line that at launch was not available in Europe.

Gemini 3.8 Flash is out with a new focus on cybersecurity

Gemini 3.8 Flash Cyber brings specialized vulnerability detection and automated patching capabilities, but access is limited to Google's Fairwind Program. Google's previous smartest AI model, Gemini 3.6 Flash, was released in late July. The tech giant wasted no time iterating on it, with Gemini 3.7 Flash launching just weeks after. Now, just three weeks later, the company is already introducing 3.7 Flash's successor. Google just announced Gemini 3.8 Flash, and this time around, the focus is on delivering next-generation intelligence for agentic workflows and cybersecurity with two different variants. The new models mark Google's third Flash release in six weeks, with new frontier models nowhere in sight. The new models are Gemini 3.8 Flash and Gemini 3.8 Flash Cyber. The former is the company's "most intelligent workhorse model," while the latter is its "most capable cybersecurity model." Gemini 3.8 Flash is primarily aimed at developers building agentic applications and workflows, though it is also available to regular Google AI Pro and Ultra subscribers via the Gemini app. According to Google's numbers, Gemini 3.8 Flash outperforms most larger frontier models when it comes to the DeepSWE v1.1 benchmark. The benchmark measures the ability to autonomously solve complex engineering problems, and Gemini 3.8 Flash hit 73.7% on it. That's just below Claude Opus 5's 74% score, and well above Gemini 3.7 Flash, Claude Sonnet 5, GPT-5.6 Sol and GPT-5.6 Terra. Gemini 3.8 Flash also outperformed all of the aforementioned models on several other benchmarks, including financial analysis on Vals Finance Agent v2, complex legal workflows on Harvey's Legal Agent Benchmark, information synthesis on CharXiv Reasoning, agentic terminal coding on Terminal-bench 2.1, long video understanding on LVBench, and more. Google explained that "3.8 Flash works harder," and hence the performance gains. On complex tasks, it reportedly exhibits greater diligence, with the model executing extra reasoning steps whenever necessary. That also means the model might, at times, use more tokens to maximize performance. Gemini 3.8 Flash launches at $0.75 per million input tokens and $3.75 per million output tokens. This is introductory pricing, which will be revised to $1.50 and $7.50, respectively. Regardless, the model should still be cheaper than counterparts like Claude Opus 5 and GPT-5.6 Sol. Gemini 3.8 Flash Cyber, on the other hand, isn't available to all. Access to it is limited through a new Fairwind Program, which will let trusted government authorities, as well as critical infrastructure operators and software maintainers leverage the new model. On the industry-standard benchmark for finding vulnerabilities in C/C++, the model scored 86.2%, beating its predecessor Gemini 3.5 Flash Cyber and larger models like Mythos 5 and GPT-5.5-Cyber. Similarly, in real-world vulnerability discovery spanning over 20 programming languages, the model reached a success rate of 71%, which is significantly higher than Gemini 3.7 Flash's 58.9% and Gemini 3.5 Flash Cyber's 46.6%. Elsewhere, what makes the model special is its automated patching ability. Gemini 3.8 Flash Cyber equips defenders with the ability to generate patches to address vulnerabilities. According to the company's numbers, the model produced 2.6 times more correct patches to vulnerabilities in Chrome "than the best commercial models that are much larger." For what it's worth, most regular users will never really interact with the new model, but the improvements it offers to defenders and developers can have an indirect impact on making your everyday tech safer and better. Gemini 3.8 Flash is rolling out now for developers, Enterprise customers, and Google AI Pro and Ultra subscribers. Access to Gemini 3.8 Flash Cyber, as mentioned earlier, is limited to Fairwind Program members.

Right on Cue, Google Is Reportedly on the Verge of Shipping a Coding-First AI Model

According to the popular narrative right now, Google is behind when it comes to AI models. In the popular imagination, it finished at the top of the pile last year, and then got steamrolled by "agents" when platforms like OpenClaw made token-guzzling agentic coding into the AI trend of 2026. So if you've been waiting for Google DeepMind to pivot to coding, like OpenAI did in March, that's now on the verge of happening, according to an anonymously-sourced story in the Wall Street Journal. The Journal says Gemini 3.8 Flash, also known as "Skimaki" could be released as early as Wednesday (It's Tuesday as I'm writing this). The anonymous insiders who spoke to the Journal touted the model's "significantly upgraded coding capabilities." To be clear, this supposed release will not be the sort of multi-trillion parameter model like OpenAI's still-unreleased Astra. Frontier AI models now routinely come with prophecies of doom before they're released, and have to endure a mysterious vetting procedure from the White House. Instead of being the biggest and strongest, Gemini Flash models, like this rumored 3.8 Flash release, are designed to run faster and cheaper, which are also increasingly valuable attributes -- thus the rise of platforms for efficiently aggregating tokens from different models, like OpenRouter. But the timing here is no accident. DeepMind founder and former boss Demis Hassabis stepped aside as Deepmind's CEO early last month. While insiders were quick to claim that he'd had one foot out the door for about a year, it was also widely reported that Google founder Sergey Brin was taking a more active interest in AI, and planning on quickly shipping more competitive Gemini models. Hassabis, remember, is a Nobel prize winner in chemistry and has sounded as recently as last year like he was in no hurry to release revolutionary models given the risks. His new role at DeepMind is largely focused on scientific research, where he seems more comfortable. The Journal's Google insiders say "Skimaki" has been tested on an internal coding tool called "Jetski," and apparently coders say they like it better than Anthropic's Claude Opus.

Gemini 3.8 Flash rolling out three weeks after last release

After the last model release three weeks ago, Google today is rolling out Gemini 3.8 Flash. Our most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows. 3.8 Flash beats its predecessor in various benchmarks: Like the previous model, Google is offering an introductory price of $0.75/1M input tokens and $3.75/1M output tokens until December 31. Gemini 3.8 Flash is already live in the Gemini app for Google AI subscribers, Antigravity, and AI Studio.

Gemini 3.8 Flash could land any day now, and it could put the vibe back into vibe coding

Google has internally scrapped plans to launch Gemini 3.5 Pro, and could only release the next Pro model with Gemini 4.0. With such a wide range of applications, Gemini feels more like a general-purpose technology or GPT (no, not that GPT) and less like a specialized frontier model. But it has made reasonable strides in vibe-coding. Earlier in March, Google CEO Sundar Pichai himself said that 75% of Google's code is vibe-coded and approved by engineers. While that could be why Pixel's software updates remain a mess, Google is continuing to advance its vibe-coding focus and is reportedly testing a new model that could compete with stalwarts such as Claude Cowork (powered by Fable or Mythos), (now) xAI-owned Cursor, and ChatGPT's Codex. DeepMind, Google's frontier AI research arm, could soon release a new model with significantly improved coding capabilities, the Wall Street Journal has reported (paywalled). Based on internal testing, Google's engineers reportedly found the new model, Gemini 3.8 Flash, better than Anthropic's Opus (unspecified version). The WSJ report further claims that Google could release Gemini 3.8 Flash as soon as later today, i.e., on September 2. This comes just weeks after Google released Gemini 3.7 Flash with coding capabilities better than Claude's Sonnet 5 (not the more weighted Opus). This would be DeepMind's second major release since its latest rejig, where co-founder, CEO, and Nobel laureate Demis Hassabis was moved to a broader role overseeing AI development across parent company Alphabet. The need to stress vibe-coding could also arise from the recent release of z.AI's GLM-5.3-Flash, which recently took the internet by storm after being previewed as a stealth model (under codename Ox Alpha) on OpenRouter. According to BencLM's vibe-coding leaderboard, Anthropic's Claude Fable and OpenAI's GPT-5.6 Sol are among the top 10 best-performing frontier models for coding. Gemini 3.7 Flash, on the other hand, ranks 17th. Meanwhile, Gemini 3.5 Pro has reportedly been scrapped internally due to insufficient upgrades over the Flash models, and Google might jump straight to Gemini 4.0 Pro, though the model reportedly seems far from complete.

Google Gemini 3.8 Flash coding AI model expected this week

Gemini 3.8 Flash, known internally as "Skimaki," is expected to release Wednesday, according to the Wall Street Journal Google $GOOGL's AI research unit plans to ship a new model boasting sharply improved coding abilities as soon as Wednesday, the Wall Street Journal reported. The model, Gemini 3.8 Flash, is referred to internally as "Skimaki" and is designed to close ground on rivals Anthropic and OpenAI in an area where Google has lagged. Engineers at Google have tested Skimaki against Anthropic's Opus model using Jetski, Google's internal coding tool, and have preferred the new model, according to the Journal. The release comes after Google has devoted more resources to reinforcement learning -- a later stage of model training that teaches models to perform skills through trial and error -- since the start of the year. A strong showing by Gemini 3.8 Flash would not by itself reestablish Google's position at the frontier of AI development. The Flash series is built to be smaller, cheaper, and faster to run than Google's flagship models, which are built from trillions of numerical parameters. The release follows leadership changes at Google DeepMind. Demis Hassabis, the unit's co-founder and Nobel Prize winner in chemistry, stepped aside as chief executive last month. His successor, Koray Kavukcuoglu, has told employees he wants to increase the pace of execution. Kavukcuoglu had already taken charge of day-to-day Gemini decisions for at least a year before his appointment, while Hassabis directed his attention toward outside commitments. Google has also struggled with its larger model lineup. Internal candidates for Gemini 3.5 Pro were scrapped because they did not represent a sufficient improvement over the Flash series. Chief Executive Sundar Pichai had said in May that a Pro model would arrive "next month," but none has materialized. Gemini 4, Google's next planned flagship, posted encouraging numbers in pretraining evaluations but has yet to finish the posttraining phase. Google introduced Gemini 3.7 Flash roughly three weeks before this latest release, marketing it to businesses developing autonomous AI systems as a lower-cost option. That model showed gains on coding benchmarks including FrontierCode 1.1 and DeepSWE v1.1 compared with its predecessor. Google also brought on Barret Zoph, who previously co-founded Thinking Machines Lab and served as OpenAI's posttraining lead, to fill a vice president of research role covering reinforcement learning and posttraining.