Gemini 3.8 Flash Cyber

Google · Released Sep 2026

Conditional

Gemini 3.8 Flash Cyber is Google's most capable cybersecurity model, a defensive specialist already in production at Chrome Security, Cloud Vulnerability Research and partner Wiz, with frontier vulnerability-discovery and patching results at a fraction of the leaders' rollout cost. The deciding caveat is access: it is gated to vetted defenders through the Fairwind Program, with no public API or published rate card. We rate it conditional rather than caution like the sibling Gemini 3.5 Flash Cyber because 3.8 ships published benchmark figures and named production deployments where 3.5 had neither, even though both remain gated. Right for organizations that can qualify as trusted defenders; everyone else cannot buy it today.

Is it right for you?

Good for

Autonomous vulnerability discovery — CyberGym Pass@1 86.2%, ahead of GPT-5.5-Cyber (85.6%), Mythos 5 (83.8%) and the prior 3.5 Flash Cyber (77.5%)
Defensive patching at scale — CWE-Bench pass@1 47.2% versus the 47.8% leader, at $3.64 mean cost per rollout against $10.27 for that leader
Real-world multilingual codebases — over 70% recall on Google's internal 20-language vulnerability set, against 46.6% for 3.5 Flash Cyber
In-production defenders — Chrome Security measured 2.6x more correct patches than leading commercial alternatives; Wiz saw +7.5-9.7% recall at 2.3-5.2x lower cost
Prompt-injection-hardened security agents — 6.0% ASR@15 on Gray Swan indirect prompt injection, close to the general 3.8 Flash at 5.5%

Not good for

Anyone without Fairwind approval — vetting is limited to government authorities, critical-infrastructure operators and software maintainers; there is no public API or self-service access
Offensive exploitation work — Google states it prioritized patching over exploitation, with cyber-offense mitigations applied under the Frontier Safety Framework
Independent verification — no separate public model card or rate card; CyberGym is Google-run in an internal Antigravity harness and the 20-language set is private

How it performs by task

Vulnerability discovery (CyberGym)

Excellent

86.2% Pass@1, top of Google's five-model comparison — 0.6pp over GPT-5.5-Cyber, though harnesses differ across owners

Audit and patching (CWE-Bench v0)

Very Good

47.2% Pass@1 at $3.64 mean cost per rollout — 0.6pp behind the leader at roughly 65% lower cost

Real-world vulnerability discovery (20-language internal set)

Very Good

Over 70% recall versus 46.6% for 3.5 Flash Cyber; private set, not externally reproducible

Prompt-injection robustness (Gray Swan)

Very Good

6.0% ASR@15 — lower is better; near Claude Opus 5's 4.8%

Offensive exploitation

Fair

Deliberately deprioritized — Google prioritized patching over exploitation

Pricing

Input

N/A (no public rate card)

Output

N/A (no public rate card)

Context

Fairwind-gated; no public API

View full pricing

Benchmarks

Benchmark	Score	Source
CyberGym Pass@1	86.2%	Source
CWE-Bench v0 Pass@1	47.2%	Source
CWE-Bench v0 mean cost per rollout	$3.64	Source
Google internal 20-language vulnerability set (recall)	>70%	Source
Gray Swan indirect prompt injection (ASR@15)	6.0%	Source
Chrome Security correct-patch rate vs leading commercial alternatives	2.6x	Source

No verdict changes yet

The clock starts day one — changes land here as our verdict evolves.

Sources

Verification log

No verification checks yet

We haven't logged a verification check for this entry. Once a check runs, its history shows here.

How we evaluate