{"id":14300,"date":"2026-09-04T06:00:35","date_gmt":"2026-09-04T06:00:35","guid":{"rendered":"https:\/\/krikis.lt\/google-gemini-3-8-flash-cyber-valdzios-institucijoms\/"},"modified":"2026-09-04T06:00:35","modified_gmt":"2026-09-04T06:00:35","slug":"google-gemini-3-8-flash-cyber-valdzios-institucijoms","status":"publish","type":"post","link":"https:\/\/krikis.lt\/en\/google-gemini-3-8-flash-cyber-government-agencies\/","title":{"rendered":"Google launches Gemini 3.8 Flash and Cyber edition for government agencies"},"content":{"rendered":"

Google has once again introduced a new version of its Flash series of large language models (LLMs) \u2013 Gemini 3.8 Flash \u2013 along with Gemini 3.8 Flash Cyber, which is designed specifically for cyber research. In today\u2019s announcement, published by The Next Web<\/em> and SiliconANGLE<\/em>, it is emphasized that this is the third Flash model released in six weeks, and that the Cyber variant is only available to trusted testers and governments.<\/p>\n

Dual model line: general and cyber version<\/h2>\n
\"Data<\/figure>\n

\u201eThe Flash series remains part of Google's strategy to offer faster and cheaper LLM solutions than the Pro line, which hasn't been updated since early 2026. The Next Web<\/em> indicates that Gemini 3.8 Flash is a "workhorse" - a general-purpose model, while Gemini 3.8 Flash Cyber is designed for vulnerability detection and repair. SiliconANGLE<\/em> confirms that both models are based on the same technical platform, but the "Cyber" variant is optimized for cyber research and is available through a new early access initiative - the Fairwind Program.<\/p>\n

Operational efficiency and productivity<\/h2>\n
\"People<\/figure>\n

Google says that Gemini 3.8 Flash performed well in 16 AI benchmark tests, including Terminal-Bench 1.1, which is designed to solve multi-step coding tasks. The model outperformed Claude Opus 5 and GPT 5.6 Sol in nine of the tests, and achieved a score of 73.7 % on DeepSWE-1.1, which is 1 % more than GPT-5.6 Sol and just a few hundred percent less than Opus 5. SiliconANGLE<\/em> points out that these gains stem from the principle of \u201emore work\u201c: the model uses additional cycles of causal steps and tool calls, sometimes consuming more tokens, to improve accuracy.<\/p>\n

The cyber variant, Gemini 3.8 Flash Cyber, is also showing strong results. SiliconANGLE<\/em> reports that the Cyber model scored 86.2 % on the CyberGym benchmark, which measures the ability to detect vulnerabilities in the C and C++ languages. This score beats Claude Mythos 5 (83.8 %) and GPT-5.6 Sol (83.6 %). Google also says that internally, the company's Chrome team saw 2.6 times more correct fixes, and another project on the team found a "critical core vulnerability" that took months to fix but was discovered in two hours.<\/p>\n

Access and pricing<\/h2>\n

\u201eFlash models remain available at an introductory price until the end of the year: $0.75 per million input tokens and $3.75 per million output tokens. After this period, prices are expected to rise to $1.50 and $7.50, respectively. This pricing is competitive, especially considering that competitors are constantly lowering token prices in an effort to keep cautious companies in business.<\/p>\n

\u201eThe Cyber option is limited to trusted testers and governments, but sources do not disclose which specific countries are receiving access. The Next Web<\/em> stresses that this restriction is linked to the requirements of EU AI law, which requires technical documentation, a copyright policy and a summary of training data for each shared model under Article 53 of 2026. Google has already joined the Common AI Code of Conduct (30 July 2025) and models with training computations exceeding 10^25 FLOPs must be notified to the Commission within two weeks under Article 52.<\/p>\n

European regulation and future prospects<\/h2>\n

Each Flash model release in Europe becomes a separate compliance event. Although the Flash series was not initially available in the European market, Google has voluntarily committed to complying with the requirements of the EU AI law. This means that each model must be documented, and its level of risk assessed and reported to regulators. This practice demonstrates Google\u2019s commitment to maintaining transparency and meeting strict regulatory standards, even if it may slow down the pace of model releases.<\/p>\n

Conclusion<\/h2>\n

Google Gemini 3.8 Flash and its Cyber variant are important steps in the evolution of large language models. The general Flash version provides a faster and more cost-effective solution for a wide range of tasks, while the Cyber variant opens up opportunities for governments and critical infrastructures to detect and fix software vulnerabilities more quickly. Regulatory requirements in Europe, especially the EU AI law, make these models not only technologically important, but also legally important. In the future, we can expect Google to continue to expand the Flash line, aiming to balance performance, cost, and compliance with strict regulatory standards.<\/p>\n

Sources<\/h2>\n