Overview
Gemini 4 Argon Cyber Permissive is the defensive-security version of Gemini 4 Argon that Google made generally available on Gemini Enterprise Agent Platform on 13 October 2026, the same day as the public gemini-4-argon model. Google's docs describe it as "a version of Gemini 4 Argon for defensive cybersecurity use cases" — vulnerability discovery, reverse engineering, exploit verification and patch generation — and classify it as a Google Frontier Cyber Model under Section 31 of the Google Cloud Service Specific Terms.
It is the API form of the release plan Google set out in its 30 September 2026 Gemini 4 Argon announcement: the general model refuses harmful requests such as cyber attacks, while for trusted defenders and Google's internal teams Google said it would release Argon "without cyber guardrails" so they can use its full cybersecurity defence capabilities. Access is allowlist-only: customers contact their Google team or apply through Google DeepMind's Fairwind Program, and Google notes that access to Gemini 3.8 Flash Cyber does not carry over — a separate application is required.
Technically it tracks gemini-4-argon with a few documented differences. It defaults to the HIGH thinking level (LOW, MEDIUM and HIGH are all supported), does not support computer use, and runs in the global and us regions. Like gemini-4-argon it has a 1,048,576-token limit shared by input and output, reachable through long decodes. It can also be used inside CodeMender, Google's code-security agent, under Google's pre-GA terms. Google has published no benchmark results for this variant separately from Gemini 4 Argon.
| Released | 2026-10-13 |
|---|---|
| License | Proprietary |
| Weights | API only |
| Parameters | Undisclosed |
| Context | 1M |
| Max output | 1M (long decodes) |
| Architecture | A version of Gemini 4 Argon for defensive cybersecurity use cases. Google's docs list the differences from the public model: allowlist-only access, HIGH as the default thinking level (MEDIUM on gemini-4-argon), no computer use, and the global and us regions only (gemini-4-argon also serves eu). |
| Modalities | Text, Vision, Audio, Video |
| Status | Generally available since 13 October 2026 on Gemini Enterprise Agent Platform, behind an allowlist — request access through your Google team or the Fairwind Program. Usable with CodeMender under preview terms. Not on the public Gemini API. |
Pricing
| Input | $4.00 / 1M tokens |
|---|---|
| Cached input | $0.20 / 1M tokens |
| Output | $20.00 / 1M tokens |
Standard on-demand price, global endpoint; non-global endpoints are $4.40 input and $22 output. Google states the Gemini 4 Argon promotional pricing (50% credits back through 30 November 2026) does not apply to Cyber Permissive. CodeMender charges the same $4 / $0.20 / $20 for this model.
Strengths
- Cyber refusals relaxed for vetted defenders, per Google's plan to release Argon without cyber guardrails to trusted defenders
- Built for vulnerability discovery, reverse engineering, exploit verification and patch generation
- Defaults to HIGH thinking, the level Google recommends for vulnerability discovery and multi-file engineering
- 1,048,576-token limit shared by input and output, with long decodes for very long responses
- Available directly on Gemini Enterprise Agent Platform or through CodeMender
Best for
- Reach for it for authorised defensive security work — finding, verifying and patching vulnerabilities — when the public Gemini 4 Argon refuses the request
- Reach for it for reverse engineering and exploit verification in vetted security programmes
- Reach for it inside CodeMender to automate vulnerability fixes with Argon-class reasoning
- Not for agents that drive a computer: computer use is not supported, unlike gemini-4-argon
- Not for EU-region deployments: it runs in the global and us regions only
How to access
| Provider | Model ID |
|---|---|
| Gemini Enterprise Agent Platform (allowlist) ↗ | gemini-4-argon-cyber-permissive |
Gemini Cyber — every version
The full lineage of the Gemini Cyber line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.
| Version | Released | Context | License |
|---|---|---|---|
| Gemini 4 Argon Cyber Permissive | 2026-10-13 | 1M | Proprietary |
| Gemini 3.8 Flash Cybercurrent | 2026-09-02 | — | Proprietary |
| Gemini 3.5 Flash Cyber | 2026-07-21 | — | Proprietary |
FAQ
What is Gemini 4 Argon Cyber Permissive?
It is a version of Gemini 4 Argon for defensive cybersecurity — vulnerability discovery, reverse engineering, exploit verification and patch generation — made generally available on Gemini Enterprise Agent Platform on 13 October 2026 behind the Fairwind Program allowlist. Its model id is gemini-4-argon-cyber-permissive.
How is it different from Gemini 4 Argon?
The public Gemini 4 Argon refuses harmful requests such as cyber attacks; Google said trusted defenders would get Argon without cyber guardrails, and Cyber Permissive is that version. Google's docs also list access (allowlist only vs publicly available), default thinking level (HIGH vs MEDIUM), computer use (not supported vs supported) and regions (global and us vs global, us and eu).
How do I get access?
Contact your Google team or apply through Google DeepMind's Fairwind Program. The Fairwind Program prioritises governments and national cyber authorities, critical infrastructure operators and maintainers of core technology platforms, and vets applicants for security history and ethical operations. Access to Gemini 3.8 Flash Cyber does not carry over — you must apply separately.
How much does Gemini 4 Argon Cyber Permissive cost?
On Google's Agent Platform pricing page it costs $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens on the global endpoint ($4.40 and $22 on non-global endpoints). Google states that the Gemini 4 Argon promotional pricing does not apply to Cyber Permissive. Through CodeMender it is also $4 / $0.20 / $20.
Does Google publish benchmarks for this variant?
No. Google's published Gemini 4 Argon results, such as its CWE-bench v1 score, are for Gemini 4 Argon; Google has not released separate benchmark figures for gemini-4-argon-cyber-permissive.