# GPT-6 Cyber, explained

> OpenAI's cybersecurity model (Nov 5, 2026): finds software flaws, builds working exploits, and writes fixes. Only approved Daybreak Red defenders can use it. API price $20/$100 per million tokens, ExploitGym 48.5%, Critical rating. Independent explainer, not affiliated with OpenAI. Updated 2026-11-07. Full page: https://lab.awaara.cn/gpt-6-cyber/

## What it is

- GPT-6 Cyber is a cyber-specialized version of GPT-6 Astra, OpenAI's general-purpose frontier model. OpenAI trained it with reinforcement learning on tasks like reproducing crashes, writing working exploits, and searching codebases for bugs, plus reward signals that cut refusals on authorized requests.
- Announced November 5, 2026. Runs through the API and Codex only. No ChatGPT version, no public signup, no published weight release.
- Rated Critical in cybersecurity under OpenAI's Preparedness Framework, up from High for GPT-5.6 Cyber. First Cyber model at Critical.

## Access (Daybreak tiers)

- Standard: no access to Cyber models.
- Daybreak Blue: defensive work on mainline models, no access to Cyber models. OpenAI calls Blue the starting point for most teams.
- Daybreak Red: GPT-5.6 Cyber for all Red organizations, GPT-6 Cyber only with a separate model approval. Requires single sign-on, multi-factor authentication, managed devices, least-privilege access, and IP allowlisting on by default for the API.
- GPT-6 Cyber is initially available to a select group of approved Daybreak Red customers, expanding to all Red customers soon (no date given). Individuals are limited to invited applicants with identity and liveness checks and a physical FIDO2 hardware key.
- Approval alone does not turn reduced refusals on: Daybreak must be enabled for each request.
- Daybreak Red for Critical Infrastructure is a new group for approved defenders testing industrial systems. The $1 billion Daybreak for Frontline Defenders program subsidizes access for resource-constrained teams.

## Benchmarks (OpenAI-reported, no independent benchmark yet)

- ExploitGym (exploit a named vulnerability in a remote target): GPT-6 Cyber 48.5%, GPT-6 Astra on Daybreak Blue 42.4%, GPT-5.6 Cyber 32.1%, GPT-5.6 on Blue 30.3%, GPT-6 Astra with standard safeguards 0.0% (declines the task).
- Blind web pentesting (no source code): 40.3% recall vs 32.0% for Astra.
- RemoteExploitBench (GitLab, PostgreSQL, Jenkins in a controlled lab): Astra on Blue refused 100% of tasks, GPT-6 Cyber completed 95%.
- Scope discipline (ExploitGym Honeypot, out-of-scope attacks without deployment instructions): GPT-6 Cyber 0.07%, GPT-5.6 Cyber 5.51%, GPT-5.6 Sol 49.5%, Astra 0%. With deployment instructions, Cyber stayed in bounds 100% of the time.
- Where it trails Astra: general coding ExpertSWE 82.95% vs 83.24%; FrontierCyber 90 of 288 vs 96; CyScenarioBench 11 of 11 solved (61% average) vs 8; blocker deception 20.29% vs 11.66%; stops when warned 73.00% vs 78.38%; human red-teaming on Codex 0.813 vs 0.977. No fully hardened target fell to either model; neither solved any of the 15 highest-difficulty challenges.

## What it has already found

- Firefox: 81 novel high-severity vulnerabilities, plus 30 known issues shown to be more severe than first reported. Mozilla published fixes in Firefox 159 (MFSA 2026-118, October 27, 2026).
- Linux kernel: reproduced 1,724 vulnerabilities, working exploits for 948 (more than 5x the 166 GPT-5.6 Sol managed with the same harness). 304 critical findings disclosed to maintainers with candidate patches. Run cost $1,213,430 in API credits, about $1,280 per working exploit.
- Brotli: critical RCE vulnerability validated with GPT-6 Cyber alongside other OpenAI models.
- Partner quotes are vendor-published testimony, not benchmarks: CodeRabbit 66% more high and medium severity issues on an internal benchmark; Cognition 23% on an internal discovery benchmark vs a previous best of 13%.

## Price (standard short-context API rates, USD, checked Nov 5-6, 2026)

- GPT-6 Cyber: $20.00 input, $100.00 output, $2.00 cached input per 1M tokens. Cache write $25 (30 min) / $50 (12 hr) per a capture of OpenAI's model page.
- GPT-6 Astra: $10.00 / $50.00 / $1.00. GPT-5.6 Cyber: $12.50 / $75.00, cache price not published.
- Context window 1,050,000 tokens; max output 128,000 tokens.

## Sources

- OpenAI, GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents (Nov 5, 2026): https://openai.com/index/introducing-gpt-6-cyber/
- OpenAI Help Center, OpenAI Daybreak: Trusted Access for Cyber overview: https://help.openai.com/en/articles/20001258-openai-daybreak-trusted-access-for-cyber-overview
- Mozilla, Security Vulnerabilities fixed in Firefox 159 (MFSA 2026-118, Oct 27, 2026): https://www.mozilla.org/en-US/security/advisories/mfsa2026-118/
- eesel.ai, GPT-6 Cyber: what it does, who can get it, and the catch (Nov 6, 2026): https://www.eesel.ai/blog/gpt-6-cyber
- VentureBeat, OpenAI launches GPT-6 Cyber with stronger exploit research and higher API price (Nov 6, 2026): https://venturebeat.com/technology/openai-launches-gpt-6-cyber-with-stronger-exploit-research-and-higher-api-price-than-anthropic-and-google
- AI Era News, GPT-6 Cyber: $1,280 a Linux Exploit, Once It Has a Login (Nov 6, 2026): https://aieranews.com/gpt-6-cyber-linux-exploit-cost/