Google has just released two new Gemini models: Gemini 3.8 Flash and Gemini 3.8 Flash Cyber. Together, they bring next-generation intelligence to two areas that matter most to modern teams: agentic workflows and cybersecurity. In this post, we’ll walk through what each model does, how they perform, and how you can get started.
Takeaways 🚀
✅Smarter models for complex workflows
✅Stronger cybersecurity with faster patching.
✅Better reasoning at lower cost
✅Built for developers and defenders
✅Safer AI with stronger protections
Notably, this is Google’s third Flash release in just six weeks. It builds directly on the momentum of Gemini 3.7 Flash, which launched only three weeks earlier. Google describes Gemini 3.8 as its best reasoning and coding model yet — and it arrives at the same speed and low cost as 3.7.
Two Models, One Shared Core
To start, Gemini 3.8 comes in two variants. Both run on the same foundational intelligence, which means they share the same core strengths in coding and reasoning.
- Gemini 3.8 Flash: Google’s most intelligent workhorse model. It improves on 3.7 Flash in software engineering, agentic tasks, and multi-step reasoning across specialized fields. Better yet, it keeps the same introductory price as 3.7 Flash: $0.75 per million input tokens and $3.75 per million output tokens.
- Gemini 3.8 Flash Cyber: Google’s most capable cybersecurity model. It offers frontier-level performance in vulnerability detection and automated patching. Trusted defenders can access it through the new Fairwind Program.
Beyond that, Google trained both models with long-running agentic loops. These loops repeatedly evaluate and refine the underlying models. In addition, rigorous training in the demanding field of cybersecurity helped drive the coding and reasoning gains across both variants.
Gemini 3.8 Flash: Built for Long-Horizon Coding and Autonomous Agents
Overall, Gemini 3.8 Flash makes big gains over 3.7 Flash. In many cases, it comes close to the performance of higher-cost frontier models — yet at a fraction of the price.
Strong Results on Key Benchmarks
Here’s how it performed across several industry tests:
- DeepSWE v1.1 (Long-Horizon Software Engineering): 3.8 Flash outperforms most larger frontier models. Specifically, it solves complex engineering problems from start to finish, and it does so at a lower cost.
- Vals Finance Agent V2 and Harvey’s Legal Agent Benchmark: 3.8 Flash beats both 3.7 Flash and other frontier models. These tests measure advanced analysis and reporting in quantitative and professional fields.
- HLE-Verified: 3.8 Flash scores 54.9%. As a result, this shows it can handle multi-step reasoning across STEM, humanities, and professional fields.
Taken together, these results point to the model’s dependability — and that matters greatly for enterprises that need reliable, autonomous performance across specialized knowledge domains.
Why the Model Works Harder
So, where do these performance gains actually come from? They stem from one core design choice: 3.8 Flash works harder. On complex tasks, it shows more diligence. It runs extra reasoning steps and calls tools again and again until the job is done.
That said, this approach does have a trade-off. At higher effort levels, the model may use more tokens to reach the best result. If you care most about compute efficiency, you have two options. You can lower the effort level to reduce token use, or you can continue using Gemini 3.7 Flash. Google still fully supports 3.7 Flash for efficiency-first workloads.
What Developers Have Built
Already, early users have put 3.8 Flash to creative work. A few examples illustrate its range:
- A playable 3D game where you guide a wizard through a castle, built from a simple looping prompt in Google Antigravity.
- A fully functional DOS-style version of Google Maps, built in a single prompt, complete with locations, directions, and Street View.
- An interactive topographic map that shows real-time cross-sections and scientific explanations, using real datasets from the U.S. Geological Survey.
- An interactive 3D hardware visualizer that decomposes devices into layers you can explode and inspect.
Gemini 3.8 Flash Cyber: Expert Cyber Performance
Shifting to the cybersecurity side, Gemini 3.8 Flash Cyber targets one goal: giving defenders an edge in a complex threat environment. It brings Flash-level speed and cost, so security teams can iterate quickly. As with 3.8 Flash, trusted defenders can access it through the Fairwind Program.
Autonomous Vulnerability Discovery
CyberGym is the standard industry benchmark for finding vulnerabilities. On this test, 3.8 Flash Cyber shows frontier-level performance in autonomous vulnerability discovery. Impressively, it beats both 3.5 Flash Cyber and much larger frontier models.
Of course, real-world defense goes beyond C/C++ code. So, Google also ran an internal benchmark. This test asks the model to find a wide range of vulnerabilities across complex codebases in 20 programming languages. Here, 3.8 Flash Cyber makes a major leap and reaches a success rate above 70%.
Automated Patching
From the beginning, Google built this model to give defenders an advantage over attackers. That’s precisely why the team invested in fixing vulnerabilities from the start — and prioritized patching over offensive capabilities like exploitation.
CWE-Bench, run by Collinear, is a tough external test for patching. On this benchmark, 3.8 Flash Cyber sits on the Pareto frontier. It scores a pass@1 of 47.2%, just behind a leading frontier model at 47.8% — but importantly, it costs far less.
Real-World Impact: Securing Google’s Code
In practice, Google already uses 3.8 Flash Cyber to protect its own code. Three real-world examples show the results:
- Chrome: The Chrome Security team found that 3.8 Flash Cyber produced 2.6 times more correct patches than the best commercial models — which are much larger.
- Wiz: Wiz reported 7.5–9.7% higher recall on its internal penetration testing benchmark. On top of that, it ran at 2.3–5.2 times lower cost than other leading frontier models.
- Google Cloud: Google’s Cloud Vulnerability Research team used the model to find a critical foundational vulnerability in under two hours — a discovery that typically takes months.
Built With Safety in Mind
Throughout development, Google kept safety at the forefront, following its Frontier Safety Framework.
- Gemini 3.8 Flash includes safeguards against misuse in Chemical, Biological, Radiological, and Nuclear (CBRN) areas, as well as cyber offense. Importantly, these safeguards still allow beneficial use cases.
- Gemini 3.8 Flash Cyber ships with a more permissive set of cybersecurity mitigations. For that reason, only trusted defenders who need broader cyber capabilities can access it.
Furthermore, both models also improved their defenses against prompt injection, as measured by Gray Swan. This helps protect users from malicious prompt-injection attacks.
How to Get Started
Ready to try one of the new models? Here are the paths for each type of user:
- Developers: Build with 3.8 Flash and explore agent-first workflows in Google Antigravity. You can also start in the Gemini API through Google AI Studio and Android Studio, or generate UIs in Stitch. Check the developer docs to begin.
- Enterprises: Access 3.8 Flash in Gemini Enterprise.
- Consumers: Google AI Pro and Ultra subscribers can use 3.8 Flash in the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets.
- Cyber teams: Apply for prioritized access to Gemini 3.8 Flash Cyber through the Fairwind Program. It’s open to trusted government authorities, critical infrastructure operators, and software maintainers.
Key Takeaways
Finally, here’s a quick summary of what these releases offer:
- Gemini 3.8 Flash delivers stronger coding, reasoning, and agentic performance at the same low price as 3.7 Flash.
- Gemini 3.8 Flash Cyber brings frontier-level skill in vulnerability discovery and automated patching, at Flash speed and cost.
- Both models share the same foundational intelligence.
- Real-world tests at Chrome, Wiz, and Google Cloud show measurable gains in accuracy, cost, and speed.
- Google built in safety features and stronger protection against prompt injection.
In short, if you build software or defend systems, the next step is straightforward. Developers can try 3.8 Flash in Google AI Studio today, while security teams can apply for access to 3.8 Flash Cyber through the Fairwind Program.
