Google logo

Gemini 3.8 Flash Cyber

Google

Gemini 3.8 Flash Cyber is Google's most capable specialized cybersecurity model, designed for autonomous vulnerability discovery and automated patching. Google reports 86.2% pass@1 on CyberGym, 71.0% recall on an internal real-world vulnerability-discovery benchmark spanning more than 1,200 vulnerabilities and 20 programming languages, and 47.2% pass@1 on CWE-Bench. Its more permissive cybersecurity safeguards make it available only to trusted government authorities, critical-infrastructure operators, and software maintainers through Google's Fairwind Program; it is not publicly available through the Gemini API or ZeroEval.

Key Specifications

Parameters
-
Context
-
Release Date
September 2, 2026
Average Score
74.6%

Timeline

Key dates in the model's history
Announcement
September 2, 2026
Last Update
September 4, 2026

Technical Specifications

Parameters
-
Training Tokens
-
Knowledge Cutoff
-
Family
-
Fine-tuned from
gemini-3.8-flash
Capabilities
MultimodalZeroEval

Benchmark Results

Model performance metrics across various tests and benchmarks

Other Tests

Specialized benchmarks
CWE-Bench
Pass@1; computed by benchmark owner Collinear AI using the Antigravity agent harness with high thinking.Self-reported
47.2%
CyberGym
Pass@1 using the official Final-submission setting and an internal non-cyber-specialized Antigravity harness.Self-reported
86.2%
Google Real-world Vulnerability Discovery
Recall over more than 1,200 recent confirmed vulnerabilities across 20 programming languages; internal non-cyber-specialized Antigravity harness.Self-reported
71.0%
Gray Swan IPI Benchmark
ASR@15 on the combined transfer-only attack set without computer use; lower is better.Self-reported
94.0%

License & Metadata

License
proprietary
Announcement Date
September 2, 2026
Last Updated
September 4, 2026

Compare Gemini 3.8 Flash Cyber

All comparisons

Similar Models

All Models

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.