Google on Wednesday, Sept. 2, 2026, introduced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber — its third Flash release in six weeks after 3.7 Flash three weeks earlier — calling Gemini 3.8 its best reasoning and coding model yet at the same speed and low cost as 3.7. The announcement came in an official Google blog post by Tulsee Doshi and Raluca Ada Popa.

Google said both variants share the same foundational intelligence and were further accelerated by long-running agentic loops, with coding and reasoning gains driven in part by rigorous training in cybersecurity. Gemini 3.8 Flash is the general workhorse; Flash Cyber is a specialized cybersecurity model limited to trusted defenders through Google’s new Fairwind Program.

Gemini 3.8 Flash: coding, agents, and price

Google positioned 3.8 Flash as its most intelligent workhorse model, with significant improvements from 3.7 Flash across software engineering, agentic tasks, and multi-step reasoning in specialized domains. On DeepSWE v1.1 (long-horizon software engineering), Google said 3.8 Flash outperforms most larger frontier models at a fraction of the cost. The company also cited gains on Vals Finance Agent V2 and Harvey’s Legal Agent Benchmark, and a 54.9% score on HLE-Verified for multi-step reasoning across STEM, humanities, and professional fields.

Introductory API pricing matches 3.7 Flash: $0.75 per million input tokens and $3.75 per million output tokens. Google said that introductory price expires Dec. 31, 2026; starting Jan. 1, 2027, rates become $1.50 / $7.50 per million input/output tokens. Developers can still use lower effort levels or keep 3.7 Flash for efficiency-first workloads.

Availability, per Google: developers via the Gemini API in Google AI Studio and Android Studio, Google Antigravity, and Stitch; enterprises via Gemini Enterprise; consumers on Google AI Pro and Ultra in the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets.

Flash Cyber and Fairwind

Gemini 3.8 Flash Cyber is aimed at vulnerability detection and automated patching. Google said it shows frontier-level performance on CyberGym for autonomous vulnerability discovery, surpassing 3.5 Flash Cyber and significantly larger frontier models; on an internal multi-language vulnerability benchmark spanning 20 programming languages, Google reported a success rate exceeding 70%. On Collinear’s CWE-Bench for patching, Google put Flash Cyber on the Pareto frontier with a pass@1 of 47.2% versus 47.8% for a leading frontier model at lower cost.

Google also claimed real-world use inside the company: the Chrome Security team found 3.8 Flash Cyber produced 2.6 times more correct patches to Chrome vulnerabilities than the best much-larger commercial models; Wiz reported +7.5–9.7% higher recall on an internal penetration-testing benchmark at 2.3–5.2× lower cost; and Google’s Cloud Vulnerability Research team said the model found a critical foundational vulnerability in under two hours.

Access is through the Fairwind Program, which Google DeepMind describes as prioritized access for trusted government authorities, critical infrastructure operators, and software maintainers. Google said 3.8 Flash Cyber ships with more permissive cybersecurity mitigations than the general Flash model and is therefore restricted to vetted defenders; applicants are reviewed for ethical operations and research history.

Google said 3.8 Flash includes safeguards against misuse in CBRN and cyber-offense domains under its Frontier Safety Framework, and that Gemini 3.8 models improved prompt-injection robustness on Gray Swan’s measurements.

Dated Wednesday, Sept. 2, 2026: the dual release is Google’s pitch that Flash-class speed and price can now carry both everyday agentic coding and a locked-down cyber specialist — with Cyber staying behind Fairwind’s gate.

Sources