Pyyan / Language Models / GPT-5.6-Cyber
GPT-5.6-Cyber
Trained to find zero-days and chain exploits. Vetted applicants only.
currentOpenAIVerified 2026-09-02
Gated
Daybreak Red only
Specifications
| Context | 400Ksource · as of 2026-09-02 · low confidence |
|---|---|
| Input ($/Mtok) | Not publishedsource · as of 2026-09-02 |
| Output ($/Mtok) | Not publishedsource · as of 2026-09-02 |
| Licence | Proprietary, vetted accesssource · as of 2026-09-02 |
| Max output | Not publishedsource · as of 2026-09-02 |
| Parameters | Not publishedsource · as of 2026-09-02 |
| Released | 10 Aug 2026source · as of 2026-09-02 |
| Model ID | gpt-5.6-cybersource · as of 2026-09-02 · medium confidence |
Official links
GPT-5.6-Cyber model pageWhat matters
- Built on GPT-5.6 Sol and trained for finding zero-days and building exploit chains
- Completes 95% of exploit-chain, privilege-escalation and auth-bypass prompts, against 1.5% for Sol
- Reachable only through Daybreak Red, the applicant-vetted tier of OpenAI's defender programme
Best for authorised security research.