Google has launched Gemini 3.8 Flash alongside a specialized Gemini 3.8 Flash Cyber model, promising stronger coding, complex reasoning and cybersecurity capabilities without increasing the introductory API price.
The two artificial intelligence models were announced on Wednesday, September 2, just three weeks after Gemini 3.7 Flash. Google described Gemini 3.8 as its strongest reasoning and coding model to date within the Flash family.
The rapid release marks Google’s third Flash model introduction in six weeks, highlighting the accelerating competition to produce AI systems capable of completing longer and more complex tasks with limited human supervision.
What Gemini 3.8 Flash offers
Gemini 3.8 Flash is designed as a general-purpose model for software development, autonomous agents and multi-step professional work. Google said it delivers notable improvements over Gemini 3.7 Flash while retaining the speed and introductory pricing of the previous release.
The model costs $0.75 per one million input tokens and $3.75 per one million output tokens during the introductory pricing period. Google says those rates will remain in effect through December 31, 2026.
Beginning January 1, 2027, the price is scheduled to increase to $1.50 per one million input tokens and $7.50 per one million output tokens.
According to Google’s official announcement, Gemini 3.8 Flash uses long-running agentic loops that allow it to repeatedly evaluate its work, call tools and refine its responses.
This approach can improve results on demanding assignments, although Google cautioned that the model may consume more tokens when operating at higher reasoning-effort levels. Developers focused primarily on efficiency can select lower effort settings or continue using Gemini 3.7 Flash.
Google reported a score of 54.9% on HLE-Verified, a benchmark covering difficult multi-step questions across science, humanities and professional subjects. It also said the model recorded improvements in software engineering, finance and legal-agent evaluations.
The launch follows a wider rush among major AI companies to build models capable of independently handling coding projects. Anthropic recently introduced an updated Claude model focused heavily on programming, as covered in our report on Claude Fable 5.1 pricing, coding features and availability.
Gemini 3.8 Flash Cyber targets software vulnerabilities
Gemini 3.8 Flash Cyber is a specialized version developed for cybersecurity teams. Its capabilities include detecting vulnerabilities across large codebases and generating patches intended to correct security flaws.
Google said the Cyber model achieved a success rate exceeding 70% in an internal vulnerability-discovery evaluation spanning 20 programming languages.
On CWE-Bench, which measures whether AI systems can correctly patch software vulnerabilities, Gemini 3.8 Flash Cyber achieved a reported 47.2% pass rate on its first attempt. Google compared that with 47.8% for an unnamed leading frontier model.
The company is already using the system inside its own operations. Google said the Chrome Security team obtained 2.6 times more correct vulnerability patches from Gemini 3.8 Flash Cyber than from larger commercial models evaluated by the team.
Google’s Cloud Vulnerability Research team also reportedly used the model to identify a critical foundational vulnerability in less than two hours. The company noted that comparable research and discovery could ordinarily take considerably longer.
Unlike the standard Flash model, Gemini 3.8 Flash Cyber is not being released for unrestricted public use. Access is initially being provided through Google DeepMind’s Fairwind Program to selected government authorities, critical-infrastructure operators and software maintainers.
Google said the restricted approach is necessary because the Cyber edition has more permissive security controls than its general model. The company has prioritized defensive functions such as vulnerability discovery and patching over offensive capabilities including exploitation.
Gemini 3.8 Flash is now available to developers through the Gemini API, Google AI Studio, Android Studio and Google Antigravity. Enterprise customers can access it through Gemini Enterprise.
Consumers with Google AI Pro or Ultra subscriptions can use the model through the Gemini app, Google Search’s AI Mode and Gemini features in Google Sheets. Availability may vary by product, account and region.
This report is based on information released by Google on September 2, 2026. Benchmark results are supplied by Google and may not independently represent performance in every real-world application.
Make Swikblog your go-to source on Google for reliable updates, smart insights, and daily trends.
