What does Google's new Gemini 3.8 Flash change for developers and security teams?

Google has released two new artificial intelligence models. They are called Gemini 3.8 Flash and Gemini 3.8 Flash Cyber. The announcement arrived on September 2, 2026, just three weeks after the 3.7 Flash debut. This marks the third Flash release in six weeks.

Same price, new ceiling

Developers pay $0.75 per million input tokens and $3.75 per million output tokens. That rate matches 3.7 Flash exactly and holds until December 31, 2026. On January 1, 2027 the price doubles to $1.50 and $7.50 per million tokens respectively. The 3.7 Flash model remains fully supported for workloads where compute efficiency matters more than peak performance.

A shared foundation powers both variants. Rigorous training in the highly demanding domain of cybersecurity drove the gains. Long-running agentic loops recursively evaluate and refine the underlying models. The result is a model that works harder on complex tasks. It executes extra reasoning steps and calls tools iteratively. At higher effort levels it may consume more tokens to maximize output quality. Developers can dial effort down to minimize token overhead.

Built for long-horizon coding

Gemini 3.8 Flash targets autonomous agents and multi-step software engineering. On DeepSWE v1.1, a benchmark for long-horizon software engineering, it outperforms most larger frontier models while costing a fraction as much. The model scores 54.9 percent on HLE-Verified, a test of multi-step reasoning across STEM, humanities, and professional fields. It also beats 3.7 Flash and other frontier models on Vals Finance Agent V2 and Harvey's Legal Agent Benchmark.

Concrete examples demonstrate the capability. A single prompt in Google Antigravity produced a playable 3D game with puzzles, environmental storytelling, and textures generated by Nano Banana. Another prompt built a fully functional DOS version of Google Maps complete with locations, directions, and Street View. A third created an interactive topographic map of famous geographical sites using real datasets from the U.S. Geological Survey. Hardware Anatomy, an interactive 3D visualizer built in Google AI Studio, generates realistic Three.js renderings of physically-proportioned teardowns for hardware devices with an explosion slider.

Cybersecurity model for defenders

Trusted defenders reach Gemini 3.8 Flash Cyber through the new Fairwind Program. Government authorities, critical infrastructure operators, and software maintainers can apply for prioritized access. The model carries a more permissive set of mitigations for cybersecurity tasks, which is why distribution stays restricted.

On CyberGym, the standard industry benchmark for autonomous vulnerability discovery, 3.8 Flash Cyber surpasses both 3.5 Flash Cyber and significantly larger frontier models. An internal benchmark spanning 20 programming languages shows a success rate exceeding 70 percent. On CWE-Bench, an external patching benchmark run by Collinear, the model achieves a pass@1 of 47.2 percent. A leading frontier model scores 47.8 percent at significantly higher cost.

Real deployment inside Google already shows measurable impact. The Chrome Security team found 3.8 Flash Cyber produced 2.6 times more correct patches for Chrome vulnerabilities than the best commercial models, which are much larger. Wiz reported 7.5 to 9.7 percent higher recall on their internal penetration testing benchmark at 2.3 to 5.2 times lower cost compared to other leading frontier models. Google's Cloud Vulnerability Research team used the model to find a critical foundational vulnerability in under two hours. Research and discovery for that class of vulnerability usually takes months.

Safety framework and access

Safeguards against misuse in chemical, biological, radiological, and nuclear domains as well as cyber offense ship with both models, per Google's Frontier Safety Framework. Gemini 3.8 models also show a significant leap in prompt injection robustness as measured by Gray Swan.

Access paths differ by audience. Developers can build with 3.8 Flash in Google Antigravity or start immediately via the Gemini API through Google AI Studio and Android Studio. UI generation is available in Stitch. Enterprise customers access 3.8 Flash through Gemini Enterprise. Consumers with Google AI Pro and Ultra subscriptions get the model across the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets. The Cyber variant stays gated behind the Fairwind Program application process.

The introductory pricing window closes at year end. After that the per-token cost doubles. How many developers will stay at the higher tier versus shifting back to 3.7 Flash for efficiency-first workloads remains to be seen.