Gemini 3.8 Flash and 3.8 Flash Cyber Release

Google DeepMind has introduced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, two models designed to enhance agentic workflows and cybersecurity. Both models share a foundational intelligence core accelerated by long-running agentic loops that recursively evaluate and refine model outputs, with significant gains driven by rigorous training in cybersecurity.

Gemini 3.8 Flash: Enhanced Reasoning and Agentic Workflows

Gemini 3.8 Flash is positioned as a high-intelligence "workhorse" model that improves upon 3.7 Flash in software engineering, agentic tasks, and multi-step reasoning. It maintains the same introductory pricing as 3.7 Flash: $0.75 per million input tokens and $3.75 per million output tokens.

Performance in Software Engineering and Professional Domains

Gemini 3.8 Flash demonstrates performance approaching that of higher-cost frontier models in several key benchmarks:

  • Long-Horizon Software Engineering: On the DeepSWE v1.1 benchmark, 3.8 Flash outperforms most larger frontier models in solving complex engineering problems end-to-end.
  • Specialized Knowledge Domains: The model outperforms 3.7 Flash and other frontier models on the Vals Finance Agent V2 and Harvey's Legal Agent Benchmark.
  • Multi-step Reasoning: 3.8 Flash achieved a score of 54.9% on HLE-Verified, covering STEM, humanities, and professional fields.

Operational Logic and Efficiency

Performance gains are attributed to a core design choice where the model executes extra reasoning steps and calls tools iteratively on complex tasks. While this increased diligence can lead to higher token usage at higher effort levels, developers can opt for lower effort levels or continue using Gemini 3.7 Flash for efficiency-first workloads.

Gemini 3.8 Flash Cyber: Specialized Cybersecurity Capabilities

Gemini 3.8 Flash Cyber is a specialized model designed for vulnerability detection and automated patching. It is available exclusively to trusted defenders through the Fairwind Program.

Vulnerability Discovery

Gemini 3.8 Flash Cyber shows frontier-level performance in autonomous vulnerability discovery:

  • CyberGym Benchmark: The model surpasses both 3.5 Flash Cyber and larger frontier models.
  • Multi-language Discovery: In an internal benchmark spanning 20 programming languages, the model achieved a success rate exceeding 70%.

Automated Patching

Google DeepMind prioritized vulnerability fixing over offensive exploitation capabilities. On the CWE-Bench (run by Collinear), Gemini 3.8 Flash Cyber is on the Pareto frontier with a pass@1 of 47.2%, nearly matching a leading frontier model (47.8%) at a significantly lower cost.

Real-World Application and Impact

Gemini 3.8 Flash Cyber has been deployed in several high-impact scenarios:

  • Chrome Security: The model produced 2.6 times more correct patches for Chrome vulnerabilities than larger commercial models.
  • Wiz Penetration Testing: Wiz reported 7.5-9.7% higher recall on internal penetration testing benchmarks at 2.3-5.2x lower cost than other leading frontier models.
  • Google Cloud Vulnerability Research: The model identified a critical foundational vulnerability in under two hours, a process that typically takes months.

Safety and Robustness

Both models include safeguards against misuse in cyber offense and Chemical, Biological, Radiological, and Nuclear (CBRN) domains, aligned with the Frontier Safety Framework. Because Gemini 3.8 Flash Cyber requires more comprehensive cyber capabilities, it ships with a more permissive set of mitigations and is restricted to trusted defenders.

Additionally, Gemini 3.8 models show improved robustness against prompt injection attacks, as measured by the Gray Swan IPI Benchmark.

Availability

  • Developers: Available via Google AI Studio, Android Studio, and the Gemini API. Agent-first workflows can be explored in Google Antigravity, and UI generation in Stitch.
  • Enterprises: Available through Gemini Enterprise.
  • Consumers: Available to Google AI Pro and Ultra subscribers via the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets.
  • Cybersecurity Professionals: Access to Gemini 3.8 Flash Cyber is managed through the Fairwind Program for government authorities, critical infrastructure operators, and software maintainers.

Sources