GPT-6 Cyber pricing and who can actually use it
- Web Security
- OpenAI
- GPT-6 Cyber
- Pricing
- Written by
- Bhargav Tibadiya
- Published
- Reading time
- 5 min read
GPT-6 Cyber costs $20 per 1M input tokens and $100 per 1M output tokens on OpenAI's API, and most developers cannot use it yet. Access needs Daybreak Red approval plus a separate approval for this model[1][4]. OpenAI introduced it on November 5, 2026, and says it is initially available to a select group of approved Daybreak Red customers[2].
This note covers price, limits and the access path. It does not cover how to exploit anything, and nothing here comes from my own testing. Every figure is from OpenAI's pages, checked on November 6, 2026.
What is GPT-6 Cyber?
OpenAI describes it as a cybersecurity model trained to find and validate exploitable vulnerabilities while staying inside the scope a researcher sets[2]. The model docs frame the use as investigating a vulnerability, reproducing it in an authorized test environment, or verifying a fix[1].
It is not a general chat model. It sits in the Daybreak program, which OpenAI calls its Trusted Access for Cyber program[3].
How much does GPT-6 Cyber cost?
Per OpenAI's model page, text token prices are[1]:
- Input: $20 per 1M tokens.
- Cached input: $2 per 1M tokens.
- Cache writes: $25 per 1M tokens.
- Output: $100 per 1M tokens.
Those are standard short-context prices. The page adds that short-context cache writes cost $25 per 1M tokens for a 30 minute cache or $50 per 1M tokens for a 12 hour cache[1].
Here is an illustrative request, with made-up sizes to show the arithmetic. Say a code review sends 200,000 input tokens and gets 10,000 output tokens back. Input costs 0.2 x $20 = $4.00 and output costs 0.01 x $100 = $1.00, so $5.00. If those 200,000 input tokens were all read from cache, input drops to 0.2 x $2 = $0.40 and the request costs $1.40. Real bills depend on how much of your prompt repeats and how long reasoning runs, because the docs list reasoning token support[1].
What are the limits and features?
The model page lists a 1,050,000 token context window, a maximum of 922,000 input tokens and 128,000 maximum output tokens. It accepts text and images and returns text[1].
It supports the Responses and Chat Completions endpoints, streaming, structured outputs, function calling and prompt caching. Batch, fine-tuning, embeddings and Realtime are listed as not supported[1]. The page's standard rate limits by tier are[1]:
- Free: 3 requests per minute, 10,000 tokens per minute.
- Build: 5,000 requests per minute, 1,000,000 tokens per minute.
- Launch: 10,000 requests per minute, 4,000,000 tokens per minute.
- Grow: 15,000 requests per minute, 40,000,000 tokens per minute.
The page says these apply to approved organizations and may differ with custom limits. A "Free" tier row does not mean a free approved account; the model still needs approval first[1].
Who can actually use it?
The model docs say your organization needs Daybreak Red access and approval to use GPT-6 Cyber[1]. OpenAI's help center splits it into levels. Daybreak Blue covers defensive work on mainline models and has no access to cyber-specialized models. Daybreak Red gives all approved organizations GPT-5.6 Cyber, and GPT-6 Cyber needs separate model approval[3].
At launch, OpenAI said GPT-6 Cyber is available through the API and Codex to a select group of approved Daybreak Red customers, with access expanding to all Daybreak Red customers "soon" and no date[2].
Approval is not automatic. OpenAI reviews requests, and the help center says finishing setup and identity checks does not guarantee approval[3].
What does the setup involve?
For an organization, OpenAI's onboarding guide lists an intake form, a Know Your Business check through Persona, and an eligibility decision. After that, an API organization owner enables Daybreak on a non-default project[4]. The guide says default projects are not eligible and changes can take about 15 minutes to apply[4].
Two controls matter for GPT-6 Cyber. The guide says single sign-on must be enforced where Daybreak Red is used, and that API access requires IP allowlisting by default[4]. In the Responses API, you set `access_programs.cyber` to `daybreak_red` for both Cyber models[4].
Daybreak is meant for authorized work on systems you own or are allowed to test. OpenAI says it cannot be extended to third-party customers or customer-facing product traffic[3].
What did OpenAI report about performance?
OpenAI reports that GPT-6 Cyber scores 48.5% on ExploitGym, up from 32.1% for GPT-5.6 Cyber[2]. It also reports 99.93% of out-of-scope-service tests avoided without its deployment prompt[2].
These are OpenAI's own numbers on its own evaluations, and I did not reproduce any of them. OpenAI also classifies the model as Critical in cybersecurity under its Preparedness Framework, which is part of why access is gated[2].
Should you plan around it?
If you run a security team that already works with OpenAI, ask your account contact about Daybreak Red and GPT-6 Cyber approval, and budget with the token prices above. If you are an individual developer or a small product team, plan on not having it today. OpenAI says individual Daybreak access needs an eligible paid plan, extra account security and a hardware key[3], and that covers Daybreak, not GPT-6 Cyber approval.
FAQ
How much does GPT-6 Cyber cost?
$20 per 1M input tokens, $2 per 1M cached input tokens, $25 per 1M tokens for cache writes and $100 per 1M output tokens, per OpenAI's model page.
Can I use GPT-6 Cyber without approval?
No. The model docs say your organization needs Daybreak Red access and approval to use it.
Is GPT-6 Cyber available to everyone?
No. OpenAI said it is initially offered to a select group of approved Daybreak Red customers and will expand to all Daybreak Red customers soon, with no date given.
What is Daybreak Red?
The higher access tier in OpenAI's Daybreak program, meant for authorized penetration testing, red teaming and controlled vulnerability research. It needs separate approval and stronger verification.
Does GPT-6 Cyber support the Batch API?
The model page lists Batch as not supported. It lists Responses and Chat Completions as supported.
How big is the context window?
1,050,000 tokens, with a maximum of 922,000 input tokens and 128,000 output tokens, per the model page.
Can I put GPT-6 Cyber behind my own product?
OpenAI says Daybreak cannot be used for customer-facing applications, downstream product traffic or third-party access.
Sources
- GPT-6-Cyber model page - OpenAI API docs (opens in a new tab)
- GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents - OpenAI (opens in a new tab)
- OpenAI Daybreak - Trusted Access for Cyber overview - OpenAI Help Center (opens in a new tab)
- Enterprise Daybreak onboarding - OpenAI Help Center (opens in a new tab)
Elsewhere on this site
The LLM API cost calculator lets you try your own token counts, and the cheapest AI APIs comparison lists prices across vendors.
