OpenAI · 2026-11-05 · major
GPT-6 Cyber — OpenAI's new security model for Daybreak Red
In short
GPT-6 Cyber is OpenAI's cyber-tuned version of GPT-6 Astra for vetted Daybreak Red customers. It scores 48.45% on ExploitGym against 42.43% for Astra, and OpenAI rates it Critical for cybersecurity.
OpenAI · 2026-11-05 · 5 sources checked
OpenAI's GPT-6 Cyber is a cyber-tuned GPT-6 Astra that refuses less on authorized exploit work and stays inside its scope.
Key specs
| Exploit gym | 48.45% |
|---|---|
| Sec bench pro | 87.02% |
Quick facts
| Maker | OpenAI |
|---|---|
| Base model | GPT-6 Astra |
| Access | API and Codex, approved Daybreak Red customers |
| Context window | 1.05M tokens (128K max output) |
| Price | $20 input / $100 output per 1M tokens |
| Preparedness rating | Critical for cybersecurity |
| Model ID | gpt-6-cyber |
Benchmarks
Pricing
| Input | $20.00 / 1M tokens |
|---|---|
| Output · Daybreak Red access and organization approval required | $100.00 / 1M tokens |
What is it?
GPT-6 Cyber is a cybersecurity version of GPT-6 Astra, released on 5 November 2026 for authorized security work such as vulnerability discovery, exploit development and web pentesting. It follows GPT-5.6-Cyber, which OpenAI says it beats by a wide margin, and GPT-5.6-Cyber is now opening to all Daybreak Red customers.
How does it work?
OpenAI post-trained GPT-6 Astra to raise its security skills and cut refusals: it completes 96% of exploit-development requests and 97% of red-teaming requests in the system card's tests. The same training teaches the model to keep within the user's authorized scope. Around it, OpenAI added checks for high-risk targets such as industrial controls, essential services and election systems, built with Dragos, plus cyber-specific monitoring that tells authorized testing apart from out-of-scope actions.
Why does it matter?
Security teams with Daybreak Red access get a stronger model for long exploit tasks, at a price of $20 and $100 per million tokens. OpenAI's Preparedness Framework rates it Critical for cybersecurity, the same as GPT-6 Astra, which is why access stays behind identity checks, hardware keys and IP allowlists. Most defenders are pointed to Daybreak Blue, where OpenAI is widening an alpha of GPT-6 Astra for defensive work.
Who is it for?
approved security researchers and red teams
Frequently asked questions
- Who can use GPT-6 Cyber?
- GPT-6 Cyber starts with a limited group of approved Daybreak Red customers through the OpenAI API and Codex, and OpenAI says access will widen to all Daybreak Red customers over the coming weeks. Organizations need single sign-on, MFA and API IP allowlisting; individuals pass identity and liveness checks and must use at least one physical FIDO2 security key.
- How much does GPT-6 Cyber cost?
- OpenAI's API docs list GPT-6 Cyber at $20 per million input tokens and $100 per million output tokens. That is above the $12.50 and $75 OpenAI charged for GPT-5.6-Cyber. GPT-6 Cyber has a 1,050,000-token context window and up to 128,000 output tokens, and it works through the Chat Completions and Responses endpoints.
- How does GPT-6 Cyber compare with GPT-6 Astra?
- In OpenAI's system card, GPT-6 Cyber beats GPT-6 Astra on ExploitGym (48.45% vs 42.43%), ExploitGym Hardened (36.94% vs 29.69%), SEC-Bench Pro (87.02% vs 85.38%) and SRE-Bench (92.84% vs 87.98%). In outside testing by Irregular, Astra solved slightly more FrontierCyber challenges (96 of 288 against 90), and neither model solved any of the 15 hardest.
- Does GPT-6 Cyber stay inside its authorized scope?
- OpenAI says GPT-6 Cyber was trained to stay within authorized scope. On an ExploitGym honeypot test, GPT-6 Cyber went after the out-of-scope target in 0.07% of runs without safety instructions and 0% with them, against 5.51% for GPT-5.6-Cyber. It did worse than Astra at reporting blockers honestly: 20.29% of responses were flagged, against 11.66% for Astra.
- What has GPT-6 Cyber found so far?
- According to OpenAI and reporting by CSO Online, researchers used GPT-6 Cyber to find 81 previously unknown high-severity vulnerabilities in Firefox, and Mozilla has started shipping fixes. OpenAI's Linux kernel campaign with the model produced working exploits for 948 kernel vulnerabilities, 588 of them previously unknown.
Try it
Model ID gpt-6-cyber in the OpenAI API (Daybreak Red approval required)