What is GPT-5.6-Cyber?
GPT-5.6-Cyber is a version of OpenAI's GPT-5.6 Sol tuned for offensive security work, announced on 10 August 2026. It is only available through Daybreak Red, the stricter of two tiers in OpenAI's Daybreak trusted-access programme for cyber defenders: Daybreak Blue gives verified defenders the standard GPT-5.6 Sol with security-specific safeguards, while Red adds GPT-5.6-Cyber for vulnerability research, exploit validation and penetration testing.
Access requires identity verification, separate Red approval, account monitoring, restricted uses and legal attestations; since 1 September, hardware security keys are mandatory for individual Daybreak accounts. Individuals apply at chatgpt.com/cyber, organisations through OpenAI's enterprise form and security vendors through its partner programme. It is also on Amazon Bedrock in US East (Ohio) for customers enrolled in the programme.
On OpenAI's API it costs $12.50 per million input tokens and $75 per million output tokens for prompts up to 272K tokens, with cached input at $1.25.
How much does GPT-5.6-Cyber cost?
| Plan | Price | What you get |
|---|---|---|
| GPT-5.6-Cyber | $12.50 in / $75 out per 1M tokens | Prompts up to 272K tokens; cached input $1.25 |
| Data-residency endpoints | 10% more |
OpenAI API (Daybreak Red only): $12.50 input / $75 output per 1M tokens up to 272K input tokens, cached input $1.25; 10% more on data-residency endpoints. Also on Amazon Bedrock for enrolled customers.
Is GPT-5.6-Cyber worth it?
GPT-5.6-Cyber exists because general models refuse offensive security work. OpenAI's own figure makes the point: it completes 95% of advanced exploit-chain and privilege-escalation prompts, against 1.5% for the standard GPT-5.6 Sol. That is a measure of willingness, not of skill: it shows the model will attempt the task, not that the exploit works.
The skill evidence is thinner but real. Chrome's July release credits OpenAI's security research with CVE-2026-15903, an out-of-bounds read and write in the V8 JavaScript engine, reported on 6 July and fixed ten days later. No independent evaluation of GPT-5.6-Cyber has been published; the UK AI Security Institute has evaluated its predecessor, GPT-5.5.
The controls are the product as much as the model. Identity checks, separate approval, monitoring, legal attestations and mandatory hardware keys mean it is built for established security teams, consultancies and vendors, not curious individuals. Anthropic offers a comparable path through Claude Mythos 5.1 in its own verification programme.
It is expensive at $12.50 in and $75 out per million tokens, more than GPT-6 Astra's $10 and $50, and it is built on the older GPT-5.6 generation. For a penetration-testing firm whose analysts already spend hours on exploit validation, that price is small; for anyone outside the programme, the question does not arise.
Where it earns its place
A penetration-testing consultancy
Validate whether discovered vulnerabilities are exploitable
Apply for Daybreak Red as an organisation, then use GPT-5.6-Cyber to draft and test exploit chains inside client-authorised scopes. A 100K-token analysis with a 20K-token report costs about $2.75.
A browser or platform vendor's security team
Find memory-safety bugs before attackers do
Point it at fuzzing output and source to triage and confirm bugs, as OpenAI's own research did for the V8 flaw fixed in Chrome in July.
A security team on AWS
Use the model under existing cloud contracts
Call openai.gpt-5.6-cyber on Amazon Bedrock in US East (Ohio) after enrolling in OpenAI's trusted-access programme.
Before you commit
- Setup
- Verify identity and request access at chatgpt.com/cyber (individuals) or through OpenAI's enterprise form, then apply separately for Daybreak Red. Hardware security keys are required.
- Integrations
- OpenAI API and ChatGPT for approved accounts; Amazon Bedrock (openai.gpt-5.6-cyber) in US East (Ohio).
- Data and privacy
- OpenAI does not train on API data unless you opt in; Daybreak accounts are monitored and limited to approved uses under legal attestations.
- Cost at scale
- $12.50/$75 per million tokens up to 272K input. 1,000 analyses of 100K input and 20K output tokens cost about $2,750 uncached.
- Limits
- Approved Daybreak Red members only; prices listed for prompts up to 272K tokens. Its 95% completion figure is OpenAI's measure of refusals, not of exploit quality.
What's good
- Completes 95% of exploit and privilege-escalation tasks that Sol refuses, per OpenAI
- Credited with a V8 vulnerability fixed in Chrome in July
- Strict access controls suit regulated security teams
- Also on Amazon Bedrock for enrolled customers
What's not
- Only for approved Daybreak Red members
- No independent capability evaluation yet
- $12.50/$75 per million tokens, above GPT-6 Astra
- Built on the GPT-5.6 generation, not GPT-6
Best for
Established security teams, consultancies and vendors doing authorised vulnerability research and penetration testing who can pass OpenAI's vetting.
GPT-5.6-Cyber alternatives
GPT-5.6-Cyber: common questions
- Is GPT-5.6-Cyber worth it?
- For vetted security teams, it removes the refusals that make general models useless for exploit work; everyone else cannot get it, and its quality has not been measured independently. We rate it 4.0 out of 5.
- How much does GPT-5.6-Cyber cost?
- OpenAI API (Daybreak Red only): $12.50 input / $75 output per 1M tokens up to 272K input tokens, cached input $1.25; 10% more on data-residency endpoints. Also on Amazon Bedrock for enrolled customers.
- What are the downsides of GPT-5.6-Cyber?
- Only for approved Daybreak Red members; no independent capability evaluation yet; $12.50/$75 per million tokens, above GPT-6 Astra; built on the GPT-5.6 generation, not GPT-6.
- Who is GPT-5.6-Cyber best for?
- Established security teams, consultancies and vendors doing authorised vulnerability research and penetration testing who can pass OpenAI's vetting.
- How much does GPT-5.6-Cyber cost for a team or in production?
- $12.50/$75 per million tokens up to 272K input. 1,000 analyses of 100K input and 20K output tokens cost about $2,750 uncached.
- Is GPT-5.6-Cyber safe for company data?
- OpenAI does not train on API data unless you opt in; Daybreak accounts are monitored and limited to approved uses under legal attestations.
- What are the best GPT-5.6-Cyber alternatives?
- The closest alternatives we have reviewed are Claude Fable 5.1, GPT-6 Astra and MiMo-V2.6-Pro.
Answers checked Oct 25, 2026.
Make GPT-5.6-Cyber? Show this rating on your site
