The race to dominate artificial intelligence shows no signs of slowing down, and Google has just made its move with the launch of Gemini 3.8 Flash and Gemini 3.8 Flash Cyber , two models arriving just weeks after the release of Gemini 3.7 Flash . The Mountain View company continues with its well-defined strategy: offering top-tier capabilities in lighter, faster, and more affordable models, without sacrificing power for complex tasks.
This new generation of Flash processors not only aims to improve performance in programming and reasoning, but also expands its scope into cybersecurity. With the Cyber variant, Google intends to take a step forward in automated system protection, an area where artificial intelligence is becoming an indispensable tool for both defense and attack.
More performance without touching prices
Gemini 3.8 Flash is presented as the most intelligent Flash model to date. Its strength lies in software engineering, agentic workflows, and multi-step reasoning in specialized domains. It's not just about answering questions, but about executing long tasks, using tools iteratively, and breaking down complex problems to achieve a goal. In practice, this represents a significant leap forward compared to traditional chatbots.
Data from Google's initial internal tests reinforces this idea. In the DeepSWE v1.1 benchmark, which evaluates complex programming tasks, Gemini 3.8 Flash achieved a 73,7% success rate, surpassing the 72,7% of GPT-5.6 Sol and coming very close to the 74% of Claude Opus 5, a much more expensive model. It also excels in areas such as financial analysis, complex legal workflows, and long video compression, where its results are directly comparable to those of leading competitor models.
And here's the most interesting part: all that extra performance doesn't come with a price increase . Gemini 3.8 Flash maintains the launch fee of Gemini 3.7 Flash, that is, $0,75 per million tokens issued and $3,75 per million tokens sold. Compared to the $5 and $25 fees of Claude Opus 5 or the $4 and $20 fees of GPT-5.6 Sol, Google's offering is in a different league in terms of economic efficiency.
Gemini 3.8 Flash Cyber: AI puts itself at the service of cyber defense
The biggest news in this announcement is undoubtedly Gemini 3.8 Flash Cyber. Google describes it as its most advanced cyber defense model to date , specifically designed to detect software vulnerabilities and help generate automatic patches, similar to how AI assists in vulnerability hunting . The underlying idea is clear: to automate the security analysis process to reduce the time a flaw goes undetected, something known in the industry as time to exposure.
The results the company is reporting are compelling. In the CyberGym benchmark, the gold standard for vulnerability detection, this model achieved an 86,2% success rate, surpassing GPT-5.6 Sol (83,6%) and Claude Mythos 5 (83,8%). In an internal test covering 20 different programming languages, the success rate exceeded 71%, well above the 58,9% of Gemini 3.7 Flash. But what's most striking are the real-world performance figures: the AI found a critical vulnerability in the Google cloud in less than two hours and generated 2,6 times more successful patches for Chrome than other larger commercial models.
However, not everyone will be able to use this model. Access will be restricted through the new Fairwind program , aimed solely at trusted advocates, similar to OpenAI's Daybreak, Anthropic's Glasswing, or Microsoft's Perception. Google wants to prevent these capabilities from falling into the wrong hands.
Availability and ecosystem
Both Gemini 3.8 Flash and the Cyber variant are already integrated into the Google ecosystem. In the case of the general-purpose model, its presence extends to Google AI Studio, Android Studio, Google Antigravity, and Gemini Enterprise , as well as the Gemini app for Google AI Pro and Ultra subscribers, AI Mode in Search, and Gemini features in Google Sheets.
For developers, this integration is key. Google's strategy isn't just about publishing a model in the API, but about bringing it to the tools already used daily in programming and development. The Flash family has thus evolved from Gemini 3, which already aimed to combine advanced reasoning with low latency, to this new generation that arrives ready to set the pace in terms of efficiency and cost.
Google has also managed to ensure that the leap in capacity doesn't come with a price increase, a significant move in an increasingly competitive market. The company is focusing on mid-range models that can handle complex tasks at a much lower cost, a trend also followed by other industry giants. The countdown to the next update has already begun, but for now, Gemini 3.8 Flash and Cyber set the bar quite high.
Google's commitment to the Flash family is solidified with this dual launch, which aims to bring high-performance artificial intelligence to a wider audience while simultaneously offering specialized cybersecurity tools to those who need them most. With competitive pricing, seamless ecosystem integration, and performance that rivals top-tier models, the American company is positioning itself very favorably for the coming months.
