The Pulse
Google pairs Gemini 3.8 Flash with a restricted cyber model
Google has released Gemini 3.8 Flash for broad use while limiting its cybersecurity-focused counterpart to vetted defenders through the Fairwind Program.
AI.info Team ·
Google releases one model broadly and locks the other behind a gate
Google is making Gemini 3.8 Flash available to developers, enterprises and subscribers while restricting its cybersecurity-focused counterpart to vetted defenders. The split reflects a tension in the launch: the same underlying intelligence powers both models, but Google says the cyber version needs stronger controls because it offers more permissive capabilities for security work.
Google announced the two models on September 2, 2026, describing Gemini 3.8 Flash as its strongest Flash model for reasoning and coding. The release marks Google’s third Flash launch in six weeks and keeps the new model at the introductory price used for Gemini 3.7 Flash: $0.75 per million input tokens and $3.75 per million output tokens.
Gemini 3.8 Flash Cyber is not generally available through the Gemini API or consumer applications. Access runs through the Fairwind Program, which gives priority to government authorities, critical-infrastructure operators and software maintainers.
Gemini 3.8 Flash spends more tokens to handle longer tasks
Google says Gemini 3.8 Flash improves on its predecessor in software engineering, agentic tasks and multi-step reasoning. The model supports adjustable effort levels, allowing developers to trade latency and token use against answer quality. Google’s API documentation lists an input limit of 1,048,576 tokens and an output limit of 65,536 tokens.
The company’s published evaluations show Gemini 3.8 Flash scoring 73.7% on DeepSWE v1.1, a benchmark for long-horizon software engineering, compared with 65.3% for Gemini 3.7 Flash. It scores 61.4% on Vals Finance Agent v2, ahead of Gemini 3.7 Flash at 59.0%, and reaches 54.9% on HLE-Verified, a multidisciplinary reasoning test.
Google says the gains come partly from having the model execute additional reasoning steps and call tools repeatedly on difficult tasks. Higher effort settings can increase token consumption, while lower settings are available for workloads where cost and response time matter more.
Enterprise users get the clearest early case
Google is positioning Gemini 3.8 Flash as a production model for software development, knowledge work and autonomous agents rather than as a single-purpose chatbot. The model is available through Google AI Studio, the Gemini API, Gemini Enterprise, Google Antigravity, AI Mode in Search and the Gemini app for Google AI Pro and Ultra subscribers.
Google’s own demonstrations include a playable game created in Antigravity, a working DOS version of Google Maps and a Three.js hardware visualizer that generates layered device teardowns. Such examples show the model operating across code, interfaces and structured information, although they are demonstrations supplied by Google rather than independent tests.
“To outpace automated threats, defenders need model performance that operates at the speed of the attack.”
Charlie Sestito, Director, Office of the CTO, Palo Alto Networks
Google’s pricing table gives the model a temporary cost advantage. The introductory rates expire on December 31, 2026. Starting January 1, 2027, Google says standard pricing will be $1.50 per million input tokens and $7.50 per million output tokens.
Flash Cyber focuses on finding and fixing vulnerabilities
Gemini 3.8 Flash Cyber shares the same foundational intelligence as the standard model but is tuned for cybersecurity work. Google says it prioritizes vulnerability discovery and automated patching over offensive exploitation, and that its training included demanding security tasks intended to improve coding and reasoning more broadly.
On an internal benchmark covering complex codebases in 20 programming languages, Google says Gemini 3.8 Flash Cyber found vulnerabilities successfully in more than 70% of cases. On the external CWE-Bench patching evaluation, the model posted a 47.2% pass-at-one rate, compared with 47.8% for Fable 5, according to Google.
Google also reports several internal and partner results. Its Chrome Security team found that the model generated 2.6 times more correct vulnerability patches than larger commercial models in its comparison. Wiz recorded 7.5% to 9.7% higher recall on an internal penetration-testing benchmark at between 2.3 and 5.2 times lower cost, while Google’s Cloud Vulnerability Research team says it found a critical foundational vulnerability in less than two hours.
Fairwind limits access to the more permissive model
Google’s safety distinction is direct. Gemini 3.8 Flash includes safeguards covering cyber offense and chemical, biological, radiological and nuclear misuse. Gemini 3.8 Flash Cyber uses a more permissive set of cybersecurity mitigations, so Google is limiting it to trusted defenders that need broader security capabilities.
The Fairwind Program gives priority access to government authorities, critical-infrastructure operators and software maintainers. Participating organizations are subject to due-diligence checks and must limit access to internal cybersecurity, incident-response or penetration-testing teams, according to Google DeepMind.
For ordinary developers, the immediate product is Gemini 3.8 Flash: a general model with a 1-million-token context window, tool use and adjustable reasoning effort. For security teams, the cyber variant is available only through an application process, with Google retaining control over who can use it and for what purpose.