Overview
Gemini 3.5 Flash Cyber is Google DeepMind's cybersecurity-tuned variant of Gemini 3.5 Flash, announced on July 21, 2026 alongside the mainline Gemini 3.6 Flash and Gemini 3.5 Flash-Lite. It is a lightweight model fine-tuned to find, validate, and patch software vulnerabilities at a lower price per token than larger models, and it is designed to run inside Google's CodeMender code-security agent rather than as a general-purpose chat model.
The model is not generally available. Google is offering it "exclusively to governments and trusted partners via CodeMender" under a limited-access pilot program, citing the dual-use risk of a strong offensive-security model, and says access will expand over time. It is also available to enterprise customers through the Gemini Enterprise Agent Platform in the same restricted way.
Inside CodeMender, multiple 3.5 Flash Cyber agents run in parallel to analyze large codebases and produce combined vulnerability reports. Google says the Google Cloud Vulnerability Research team used the model to uncover remote-code-execution vulnerabilities in public APIs and to find a memory-corruption vulnerability in a sensitive production service in about two hours.
| Released | 2026-07-21 |
|---|---|
| License | Proprietary |
| Weights | API only |
| Parameters | Undisclosed |
| Architecture | Fine-tuned from Gemini 3.5 Flash for cybersecurity vulnerability discovery, validation, and patching. Deployed under Google's CodeMender agent, which invokes 3.5 Flash Cyber multiple times so agents can analyze more code paths in parallel. |
| Modalities | Text |
| Status | Limited-access pilot. Available to governments and trusted partners via CodeMender; broader access planned over time. Not on the public Gemini API. |
Benchmarks
V8 JavaScript engine vulnerability run — unique confirmed issues found (Google's own evaluation).
| Benchmark | Gemini 3.5 Flash Cyber | Gemini 3.5 Flash | Claude Opus 4.6 |
|---|---|---|---|
| V8 JavaScript engine — unique confirmed issues | 55 issues | 47 issues | 36 issues |
Strengths
- Fine-tuned for vulnerability discovery, validation, and patch generation on real codebases
- Higher recall than mainline Gemini 3.5 Flash — and than Claude Opus 4.6 — on Google's V8 JavaScript-engine evaluation
- Runs at Flash-tier cost per token, so CodeMender can invoke it many times in parallel across code paths
- Deployed inside CodeMender with restricted access to reduce misuse risk while giving defenders a head start
Best for
- Automated vulnerability discovery and validation in large open-source or enterprise codebases
- Agentic patch generation as part of Google's CodeMender pipeline
- Reach for it inside authorized government or partner security programs where a limited-access, security-tuned model is required
- Not a general-purpose chatbot: use mainline Gemini 3.5 Flash / 3.6 Flash for everything outside vulnerability workflows
FAQ
What is Gemini 3.5 Flash Cyber?
Gemini 3.5 Flash Cyber is a cybersecurity-tuned variant of Google's Gemini 3.5 Flash model, announced on July 21, 2026. It is fine-tuned to find, validate, and patch software vulnerabilities and is designed to run inside Google's CodeMender code-security agent rather than as a general-purpose chat model.
Who can use Gemini 3.5 Flash Cyber?
It is not on the public Gemini API. Google is offering the model exclusively to governments and trusted partners through a limited-access CodeMender pilot, and to enterprise customers through the Gemini Enterprise Agent Platform under similar restrictions. Broader access is planned over time.
How does Gemini 3.5 Flash Cyber compare to Claude Opus 4.6 on vulnerability discovery?
On Google's V8 JavaScript engine evaluation, Gemini 3.5 Flash Cyber found 55 unique confirmed issues, versus 47 for mainline Gemini 3.5 Flash and 36 for Claude Opus 4.6. Google also reports it reaches competitive performance on the CyberGym benchmark against significantly larger models.
Why is Gemini 3.5 Flash Cyber not generally available?
Google says a strong offensive-security model is dual-use, so it is gating access to reduce misuse risk. Deployment is restricted to governments and trusted partners so defenders get a head start on finding and fixing critical vulnerabilities before they can be exploited.