Google today announced Gemini 3.8 in two variants: Gemini 3.8 Flash and Gemini 3.8 Flash Cyber. The release follows 3.7 Flash three weeks earlier and marks the company's third Flash release in six weeks. Both variants are built on the same foundational intelligence and the company attributes performance gains to long-running agentic loops that iteratively evaluate and refine the models, as well as targeted training in demanding domains such as cybersecurity.
Purpose and pricing
- Gemini 3.8 Flash: positioned as the general-purpose, intelligent workhorse, delivering meaningful improvements over 3.7 Flash in software engineering, agentic tasks, and multi-step critical reasoning across specialized domains. Introductory pricing matches 3.7 Flash at $0.75 per million input tokens and $3.75 per million output tokens. The introductory price is effective through December 31, 2026; starting January 1, 2027, prices will rise to $1.50/1M input tokens and $7.50/1M output tokens.
- Gemini 3.8 Flash Cyber: a variant tuned for cybersecurity tasks, offered to trusted defenders via the Fairwind Program. It aims to provide Flash speed and cost with capabilities focused on rapid iteration in vulnerability discovery and automated patching.
Google notes that Gemini 3.7 Flash will remain available and supported for efficiency-first workloads, and developers can choose lower effort settings on 3.8 Flash to reduce token use when compute costs are the primary constraint.
Performance and benchmarks
- Long-horizon coding: on DeepSWE v1.1 (Long-Horizon Software Engineering), Gemini 3.8 Flash reportedly outperforms many larger frontier models on autonomous end-to-end solutions for complex engineering tasks while operating at a fraction of the cost.
- Professional and quantitative benchmarks: 3.8 Flash surpasses 3.7 Flash and other frontier models on tests such as Vals Finance Agent V2 and Harvey's Legal Agent Benchmark. The model scores 54.9% on HLE-Verified, indicating multi-step reasoning ability across STEM, humanities, and professional areas.
- Operational behavior: 3.8 Flash is designed to ‘work harder’ — on complex tasks it often performs extra reasoning steps and iteratively calls tools, which can increase token usage, especially at higher effort settings.
Demonstrations and use cases
Google presented several demonstrations built with 3.8 Flash in Google Antigravity and Google AI Studio: a wizard-themed 3D puzzle game created with a single prompt and looping instructions; a fully functional DOS-style Google Maps playable with locations, directions and Street View; topographic maps built from U.S. Geological Survey datasets with real-time cross-sections and projections; and Hardware Anatomy, an interactive 3D visualizer that generates realistic Three.js renderings of device teardowns and allows layer-by-layer inspection.
Gemini 3.8 Flash Cyber: cyber capabilities
- Autonomous vulnerability discovery: on the industry standard CyberGym benchmark, Gemini 3.8 Flash Cyber demonstrates frontier-level autonomous vulnerability discovery, outperforming 3.5 Flash Cyber and several much larger frontier models.
- Broad language coverage: in an internal benchmark spanning complex codebases across 20 programming languages, the model achieved a success rate exceeding 70%, representing a substantial improvement over earlier versions.
- Automated patching: on CWE-Bench (run by Collinear), Gemini 3.8 Flash Cyber sits on the Pareto frontier with a pass@1 of 47.2%, compared to a leading frontier model at 47.8%, while being offered at significantly lower cost.
Real-world impact and partner results
Google reports several applied results using 3.8 Flash Cyber:
- The Chrome Security team found that 3.8 Flash Cyber produced 2.6 times more correct patches for vulnerabilities in Chrome than the best commercial models that are substantially larger.
- Wiz reported that Gemini 3.8 Flash Cyber achieved +7.5–9.7% higher recall on their internal penetration testing benchmark while costing 2.3–5.2x less than leading frontier models.
- Google’s Cloud Vulnerability Research team used 3.8 Flash Cyber to find a critical foundational vulnerability in under two hours — a discovery that typically takes months of research.
Safety, mitigations and access
Gemini 3.8 models include safeguards aligned with Google’s Frontier Safety Framework for Chemical, Biological, Radiological and Nuclear (CBRN) domains and cyber offense. Gemini 3.8 Flash Cyber ships with a more permissive mitigation set for cybersecurity use cases and therefore is only available to trusted defenders through the Fairwind Program. Google also reports substantial improvements in prompt injection robustness as measured by Gray Swan.
Availability
- Developers: can build with 3.8 Flash and explore agent-first workflows in Google Antigravity or via the Gemini API in Google AI Studio and Android Studio; UIs can be generated with Stitch. Developer documentation is provided.
- Enterprises: can access 3.8 Flash through Gemini Enterprise.
- Consumers: 3.8 Flash is available to Google AI Pro and Ultra subscribers in the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets.
- Cyber partners: trusted government authorities, critical infrastructure operators and software maintainers can apply for prioritized access to Gemini 3.8 Flash Cyber via the Fairwind Program.
The introductory pricing is valid until December 31, 2026, after which token rates increase as noted above. Google frames the 3.8 Flash release as a step to deliver higher-end reasoning and coding capabilities at Flash speed and lower cost to a broader set of users.
