What is GPT-6 Cyber?
GPT-6 Cyber is a cyber-specialised variant of GPT-6 Astra, released on 5 November 2026, for authorised security work: finding vulnerabilities, writing and validating exploits and web penetration testing. OpenAI trained it with reinforcement learning on crash reproduction, exploit writing and code search, and trained it to stay inside the scope a researcher sets.
It replaces GPT-5.6-Cyber at the top of OpenAI's Daybreak trusted-access programme. At launch it is available only to selected, approved Daybreak Red customers through the API and Codex, with IP allowlisting on by default for API access; OpenAI plans to extend it to all Daybreak Red customers. Individuals need an invitation, identity and liveness checks, legal attestations and at least one physical FIDO2 security key.
The API model, gpt-6-cyber, has a 1,050,000-token context window with up to 128,000 output tokens, takes text and images and returns text. It costs $20 per million input tokens and $100 per million output tokens.
How much does GPT-6 Cyber cost?
| Plan | Price | What you get |
|---|---|---|
| gpt-6-cyber | $20 in / $100 out per 1M tokens | Cached input $2 |
| Cache writes | $25 (30 minutes) or $50 (12 hours) per 1M tokens |
OpenAI API (Daybreak Red only): $20 input / $100 output per 1M tokens, cached input $2; cache writes $25 (30 minutes) or $50 (12 hours). No long-context or Batch price listed.
Is GPT-6 Cyber worth it?
GPT-6 Cyber is a meaningful upgrade, not a rename. On ExploitGym, which asks a model to turn a known vulnerability into a working exploit, it scores 48.5% against 32.1% for GPT-5.6-Cyber and 42.43% for GPT-6 Astra; on the hardened version of the test it scores 36.94% to Astra's 29.69%. Its scope discipline improved more: in OpenAI's honeypot test it avoided exploiting out-of-scope services in 99.93% of cases without safety prompts, against 94.49% for GPT-5.6-Cyber, and 100% with them.
The field results are the stronger evidence. OpenAI says the model helped find 81 previously unknown Firefox vulnerabilities that Mozilla rated high severity, and built working exploits for 948 Linux kernel vulnerabilities. Mozilla, Palo Alto Networks, Cisco and Calif are named as early users.
The gains are narrow. It does not beat GPT-6 Astra on general coding (82.95% against 83.24% on ExpertSWE), and on FrontierCyber it solved 90 of 288 challenges to Astra's 96. The system card says no attack succeeded on fully hardened targets, neither model solved any of the 15 hardest FrontierCyber challenges, and ExploitBench may be contaminated because its bugs date back to 2023. Every figure here is OpenAI's own; no independent evaluation has been published. OpenAI rates it Critical in cybersecurity under its Preparedness Framework, the same as Astra.
At $20 in and $100 out per million tokens it costs twice GPT-6 Astra's $10 and $50 and well above GPT-5.6-Cyber's $12.50 and $75. A 100K-token analysis with a 20K-token report comes to about $4.00. For a firm already paying analysts to validate exploits by hand, that is small; the harder cost is the vetting, which rules out freelancers without an invitation and organisations without SSO, MFA and managed devices.
Where it earns its place
A browser or operating-system vendor's security team
Find memory-safety bugs before attackers do and propose patches
Run it through Codex against source and fuzzing output, the kind of work in which OpenAI's researchers found the 81 high-severity Firefox flaws Mozilla has since been fixing; Codex sandboxing and Auto-Review are OpenAI's recommended controls.
A penetration-testing consultancy with Daybreak Red approval
Check which findings in a client's scope are really exploitable
Call gpt-6-cyber through the Responses API from allowlisted IPs. A 100K-token analysis with a 20K-token report costs about $4.00, so 1,000 such runs cost about $4,000 uncached.
Before you commit
- Setup
- Organisations apply for Daybreak Red and must use SSO, MFA, managed devices and least-privilege access; individuals need an invitation, identity and liveness checks, Advanced Account Security and a physical FIDO2 key. API access is IP-allowlisted by default.
- Integrations
- OpenAI API (Chat Completions and Responses) and Codex; no Batch, Realtime or fine-tuning endpoints.
- Data and privacy
- Standard data retention is the default; Zero Data Retention customers generally must use Private Safety Processing. Use is monitored, and access can be restricted, suspended or revoked.
- Cost at scale
- $20/$100 per million tokens with $2 cached input; 1,000 analyses of 100K input and 20K output tokens cost about $4,000 uncached.
- Limits
- Context window 1,050,000 tokens with up to 922,000 input and 128,000 output; no published long-context price; extra blocking checks on industrial control, essential services and election systems.
What’s good
- ExploitGym score of 48.5%, up from 32.1% for GPT-5.6-Cyber
- Stayed in scope in 99.93% of honeypot tests without safety prompts
- Credited with 81 high-severity Firefox bugs and 948 Linux kernel exploits
- 1,050,000-token context window for large codebases
What’s not
- Only for selected, approved Daybreak Red customers at launch
- $20/$100 per million tokens, twice GPT-6 Astra
- No better than Astra at general coding, and behind it on FrontierCyber
- All benchmark results are OpenAI's own
Best for
Established security teams, penetration-testing firms and software vendors with Daybreak Red approval who validate exploits at volume.
GPT-6 Cyber alternatives
GPT-6 Cyber: common questions
- Is GPT-6 Cyber worth it?
- A real step up from GPT-5.6-Cyber on exploit work and on staying in scope, at twice GPT-6 Astra's price and behind the strictest vetting OpenAI runs. We rate it 4.3 out of 5.
- How much does GPT-6 Cyber cost?
- OpenAI API (Daybreak Red only): $20 input / $100 output per 1M tokens, cached input $2; cache writes $25 (30 minutes) or $50 (12 hours). No long-context or Batch price listed.
- What are the downsides of GPT-6 Cyber?
- Only for selected, approved Daybreak Red customers at launch; $20/$100 per million tokens, twice GPT-6 Astra; no better than Astra at general coding, and behind it on FrontierCyber; all benchmark results are OpenAI's own.
- Who is GPT-6 Cyber best for?
- Established security teams, penetration-testing firms and software vendors with Daybreak Red approval who validate exploits at volume.
- How much does GPT-6 Cyber cost for a team or in production?
- $20/$100 per million tokens with $2 cached input; 1,000 analyses of 100K input and 20K output tokens cost about $4,000 uncached.
- Is GPT-6 Cyber safe for company data?
- Standard data retention is the default; Zero Data Retention customers generally must use Private Safety Processing. Use is monitored, and access can be restricted, suspended or revoked.
- What are the best GPT-6 Cyber alternatives?
- The closest alternatives we have reviewed are GPT-5.6-Cyber, GPT-6 Astra and Claude Fable 5.1.
Answers checked Nov 6, 2026.
Make GPT-6 Cyber? Show this rating on your site
