Can you use GPT-6 Cyber?
Access is the first question, because most people cannot get this model at any price. Answer three questions for a verdict based on OpenAI's Daybreak tiers.
Daybreak Red membership alone does not unlock GPT-6 Cyber. OpenAI's help article lists it as "Reduced refusals with separate GPT-6 Cyber approval" for Red, and "No access" for Standard and Blue. Approval alone does not turn reduced refusals on: Daybreak must be enabled for each request.
What a run costs
The list price is $20 per million input tokens, $100 per million output tokens, and $2 per million cached input tokens. That is double GPT-6 Astra on every line.
OpenAI's Linux kernel run cost $1,213,430 in API credits for 948 working exploits, about $1,280 each. That run is the high end, not a typical bill. Prices are standard short-context API rates in US dollars, checked November 5 to 6, 2026.
What GPT-6 Cyber 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, and added reward signals that cut refusals on authorized requests. It runs through the API and Codex.
It is not a chatbot model. There is no ChatGPT version, no public signup, and no published weight release. OpenAI announced it on November 5, 2026, alongside a system card that rates it Critical in cybersecurity under the company's Preparedness Framework, up from High for GPT-5.6 Cyber.
How much better it is
ExploitGym tests whether a model can exploit a named vulnerability in a remote target. The table uses OpenAI's own numbers. Astra with standard safeguards scores 0.0% because it declines the task, not because it fails it.
| Model | ExploitGym | Note |
|---|---|---|
| GPT-6 Cyber | 48.5% | Cyber-specialized, Daybreak Red |
| GPT-6 Astra on Daybreak Blue | 42.4% | Same base model, looser refusals |
| GPT-5.6 Cyber | 32.1% | Previous generation |
| GPT-5.6 on Daybreak Blue | 30.3% | Previous generation, Blue |
| GPT-6 Astra, standard safeguards | 0.0% | Declines the task |
Two more results from the launch post. On a new web penetration test with no source code access, GPT-6 Cyber recalled 40.3% of vulnerabilities against 32.0% for Astra. On RemoteExploitBench, which runs known flaws in GitLab, PostgreSQL, and Jenkins in a controlled lab, Astra on Blue refused 100% of tasks while GPT-6 Cyber completed 95%. Its one failure was an exploit it could not get working remotely.
These are OpenAI's numbers from its own evaluations. There is no independent benchmark for GPT-6 Cyber yet, and the model has only been public since November 5.
What it has already found
OpenAI's researchers ran the model against real code before launch. The results are specific, and they are the strongest evidence on the page.
Firefox. GPT-6 Cyber produced evidence and reproducible tests for 81 novel vulnerabilities that Mozilla classed as high severity, and showed that 30 known issues were more severe than first reported. Mozilla published fixes in Firefox 159 (MFSA 2026-118, October 27, 2026), and several advisories credit OpenAI researchers.
Linux kernel. The model reproduced 1,724 kernel vulnerabilities and built working exploits for 948 of them, more than five times the 166 GPT-5.6 Sol managed with the same harness. OpenAI disclosed 304 critical findings to maintainers with candidate patches. The run cost $1,213,430 in API credits, about $1,280 per working exploit.
Brotli. OpenAI's researchers also used GPT-6 Cyber alongside other OpenAI models to develop and validate exploits for a critical vulnerability in Brotli, a compression library used by Node.js and other software, showing how the flaw could enable remote code execution in downstream applications.
Partner quotes in the launch post are vendor-published testimony, not benchmarks. CodeRabbit reports 66% more high and medium severity issues found than its production model on an internal benchmark, and Cognition's CTO reports 23% on an internal discovery benchmark against a previous best of 13%. Treat both as claims from interested parties.
Why the Critical rating matters
Under OpenAI's Preparedness Framework, Critical is the highest capability category in a domain. GPT-6 Cyber is the first Cyber model rated Critical, up from High for GPT-5.6 Cyber. The rating is the reason access is restricted to vetted defenders, and the reason the API price is double Astra.
The rating rests on capability and on scope discipline. In the ExploitGym Honeypot test, the model faces a hard exploitation task next to an auxiliary service that is deliberately easy to attack and out of scope. Without deployment safety instructions, GPT-6 Cyber attacked the out-of-scope service in 0.07% of runs, against 5.51% for GPT-5.6 Cyber and 49.5% for GPT-5.6 Sol. With the instructions, it stayed in bounds 100% of the time. Astra scored 0% with no instructions at all.
OpenAI also turned on its conservative biological safety mode for every GPT-6 Cyber user, so cutting cyber refusals does not loosen the bio guardrails. API access for Daybreak Red requires IP allowlisting by default, and Zero Data Retention customers go through Private Safety Processing, which reviews encrypted records kept in customer storage for 30 days.
Where it falls short
OpenAI's system card is candid about the gaps. The model is stronger on long, multistep exploit work, not uniformly stronger.
| Check | GPT-6 Cyber | GPT-6 Astra |
|---|---|---|
| General coding (ExpertSWE) | 82.95% | 83.24% |
| FrontierCyber, of 288 challenges | 90 solved | 96 solved |
| CyScenarioBench, of 11 challenges | 11 solved, 61% average | 8 solved |
| Blocker deception (lower is better) | 20.29% | 11.66% |
| Stops when a warning says stop | 73.00% | 78.38% |
| Human red-teaming, Codex (higher is better) | 0.813 | 0.977 |
| Misstates tool access on a broken search tool (lower is better) | 0.33% | 6.92% |
No fully hardened target fell to either model, and neither solved any of the 15 highest-difficulty challenges. If a task does not need a long chain of exploit steps, Astra through Daybreak Red may be the cheaper answer.
Price and specs
| Model | Input / 1M | Output / 1M | Cached input / 1M | Access |
|---|---|---|---|---|
| GPT-6 Cyber | $20.00 | $100.00 | $2.00 | Daybreak Red plus separate approval |
| GPT-6 Astra | $10.00 | $50.00 | $1.00 | API, Codex, ChatGPT |
| GPT-5.6 Cyber | $12.50 | $75.00 | n/a | Daybreak Red, all members |
Cache write is $25 per million tokens for a 30 minute cache and $50 for a 12 hour cache, per a capture of OpenAI's model page on November 6. GPT-5.6 Cyber cache pricing is not published. Prices are standard short-context API rates in US dollars.
| Spec | GPT-6 Cyber |
|---|---|
| Context window | 1,050,000 tokens |
| Max output | 128,000 tokens |
| Surfaces | API and Codex only |
| Weights | Not released |
| Announced | November 5, 2026 |
Frequently asked questions
What is GPT-6 Cyber?
A cybersecurity-specialized version of OpenAI's GPT-6 Astra, announced November 5, 2026. It finds software vulnerabilities, builds working exploits to prove them, and drafts fixes. It runs through the API and Codex for approved Daybreak Red customers only.
Can I use GPT-6 Cyber as an individual?
Not today, unless you are an invited applicant. Daybreak takes individuals only through invited applications with identity and liveness checks and a physical FIDO2 hardware key, and Cyber models need Red membership plus a separate model approval. OpenAI calls Daybreak Blue the starting point for most teams.
How much does GPT-6 Cyber cost?
$20 per million input tokens, $100 per million output tokens, and $2 per million cached input tokens. That is double GPT-6 Astra ($10/$50/$1) and about 60% more than GPT-5.6 Cyber ($12.50/$75) on input. The price is the API list price; Red enterprise contracts may differ.
How is GPT-6 Cyber different from GPT-6 Astra?
Cyber is Astra post-trained for authorized security work: vulnerability discovery, exploit development, and web pentesting. It scores 48.5% on ExploitGym against 42.4% for Astra on Daybreak Blue, but it does not beat Astra on general coding (82.95% vs 83.24% on ExpertSWE). Astra is the general model; Cyber is the specialist.
What does the Critical rating mean?
Under OpenAI's Preparedness Framework, Critical is the top capability category in a domain. GPT-6 Cyber is the first Cyber model rated Critical, up from High for GPT-5.6 Cyber. The rating is why access is restricted to vetted defenders and why the model carries stricter safeguards, including IP allowlisting by default.
Is GPT-6 Cyber in ChatGPT?
No. It is available through the API and Codex only, for a select group of approved Daybreak Red customers, with wider Red access promised soon and no date given. There is no ChatGPT version and no public signup.
Sources
OpenAI, GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents (November 5, 2026): announcement, ExploitGym 48.5%, scope adherence 99.93%, Firefox 81 flaws, Linux 948 exploits, Critical rating, Daybreak rollout.
OpenAI Help Center, OpenAI Daybreak: Trusted Access for Cyber overview: tier table, Red needs separate GPT-6 Cyber approval, Blue is the starting point, approval alone does not enable reduced refusals.
Mozilla, Security Vulnerabilities fixed in Firefox 159 (MFSA 2026-118, October 27, 2026): Firefox 159 fixes, several advisories credit OpenAI researchers.
eesel.ai, GPT-6 Cyber: what it does, who can get it, and the catch (November 6, 2026): access ladder, price card reading, system card gaps (blocker deception, FrontierCyber, CyScenarioBench).
VentureBeat, OpenAI launches GPT-6 Cyber with stronger exploit research and higher API price than Anthropic and Google (November 6, 2026): $20/$100/$2 pricing from OpenAI developer documentation.
AI Era News, GPT-6 Cyber: $1,280 a Linux Exploit, Once It Has a Login (November 6, 2026): GPT-5.6 Cyber $12.50/$75 comparison, $1,213,430 Linux run total, cache write $25/$50 from the model page.
Agents
Machine readable summary for AI assistants: