Google launches Gemini 3.8 Flash and a restricted cybersecurity variant

Both models run on the same foundational intelligence, Google said, with the principal difference being the safety envelope around each version. Gemini 3.8 Flash is generally available through the Gemini API, Google AI Studio, Antigravity, Android Studio, and Gemini Enterprise, and weighs are closed, meaning there is no self-hosting or on-premises option. Gemini 3.8 Flash Cyber is not openly deployable: access is granted case by case through a new early-access initiative called the Fairwind Program.

Gemini 3.8 Flash retains the specifications of its predecessor, including a 1,048,576-token context window, a 65,536-token maximum output, text, image, audio, and video input, and text output. Thinking levels remain LOW, MEDIUM, and HIGH, with MEDIUM as the default; however, MINIMAL is not supported on the new model and returns an API validation error, according to Google's developer guide.

Introductory pricing for Gemini 3.8 Flash matches that of Gemini 3.7 Flash: $0.75 per million input tokens and $3.75 per million output tokens through Dec. 31, 2026. Prices are scheduled to rise to $1.50 and $7.50 in 2027. Google says the new model "works harder" on complex tasks, performing extra reasoning steps and calling tools iteratively, which can increase token consumption at higher effort levels. The company advises developers who face compute-constraints to consider staying on Gemini 3.7 Flash.

Google reports benchmark improvements over both its predecessor and larger rival models. On DeepSWE v1.1, a long-horizon software-engineering benchmark, the company says Gemini 3.8 Flash outperformed most larger frontier models. Reported scores on that benchmark vary: Android Authority and SiliconANGLE both cited a Google-reported 73.7 percent, while Chinese technology outlet ifanr put the figure at 71.0 percent in its own account. Google also claims gains on HLE-Verified, Vals Finance Agent V2, Harvey's Legal Agent Benchmark, and other evaluations.

The Verge noted early third-party assessments highlighting the model's efficiency: Artificial Analysis called Gemini 3.8 Flash "the cheapest we've measured at this level of intelligence" despite unchanged per-token pricing, because the model uses about 30 percent more output tokens per task and more turns on agentic evaluations. Aigora.ai CEO John Ennis described it as offering "Opus 5 coding quality but at a fraction of the cost."

Beyond standard Flash, Google introduced Gemini 3.8 Flash Cyber, which uses the same core but is tuned for finding and fixing software vulnerabilities. Google said the model achieved 86.2 percent on CyberGym, an industry-standard benchmark for detecting vulnerabilities in C and C++ code, beating its predecessor and larger models. On an internal benchmark spanning 20 programming languages, the discovery success rate was above 70 percent. For patching, Google cited Collinear's CWE-Bench result of 47.2 percent pass@1, against a leading frontier model's 47.8 percent. Google also said Chrome security teams produced 2.6 times more correct patches with the model than with larger commercial models.

Access to Gemini 3.8 Flash Cyber is limited to vetted defenders—government agencies, critical-infrastructure operators, and software maintainers—through the Fairwind Program. Google said the program launched with more than 650 participants, including Snowflake, CrowdStrike, Datadog, and the Center for Internet Security. Participants can also use Google's CodeMender harness, which assists with vulnerability discovery, severity assessment, and patch generation. Google said its own Cloud Vulnerability Research team used Flash Cyber to find a critical foundational vulnerability in under two hours, work that would normally take months.