# Is GPT-6 Cyber SOTA?

Pending benchmarks. Announced Nov 5, 2026; no independent results yet.

As of Nov 22, 2026, GPT-6 Cyber is pending benchmarks on isitsota: announced Nov 5, 2026, with no independent results on the boards we track yet.

OpenAI's cybersecurity agent model for finding, validating and helping fix exploitable vulnerabilities while staying inside the scope researchers set. Successor to GPT-5.6 Cyber. Initially available through the API and Codex to a select group of approved Daybreak Red customers, with access expanding over time; GPT-5.6 Cyber is being opened to all Daybreak Red customers. Daybreak Blue alpha access for GPT-6 Astra was expanded the same day (Astra is not a new page here). No open weights.

Canonical: https://isitsota.com/m/gpt-6-cyber
Last updated: 2026-11-05
Tracking since: 2026-11-05

## Announced facts

| Fact | Value |
| --- | --- |
| Lab | OpenAI |
| Announced | Nov 5, 2026 |
| API | Waitlist |
| Open weights | Closed · Not released |
| Role | Cybersecurity research agent: find and validate exploitable vulnerabilities, develop working exploits in authorized settings, and help produce candidate patches |
| Access | API and Codex for a select group of approved Daybreak Red customers at launch; OpenAI says access will expand to all Daybreak Red customers. New users apply via Daybreak |
| Predecessor | Successor to GPT-5.6 Cyber (OpenAI); separate from GPT-6 Rosalind, GPT-6 Astra and GPT-6 Astra Law |
| Preparedness (OpenAI) | Classified Critical in cybersecurity under OpenAI's Preparedness Framework (up from High for GPT-5.6 Cyber) |
| Alignment focus | Specialized training to stay within authorized scope, respect restrictions and report results accurately (ExploitGym Honeypot and related evals in the system card) |
| Related Daybreak | Daybreak Red hosts GPT-6 Cyber for deeper exploit development; Daybreak Blue alpha for GPT-6 Astra expanded the same day for verified defenders |
| API model ID | Not stated on the Nov 5 blog beyond availability through the API and Codex |
| Size | Not stated |

Tags marked unsure are not stated on the lab's own pages.

## What would let us call it

Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.

- [x] Publicly announced with a source
- [ ] Usable by the public (API or weights)
- [ ] Listed on an independent board we track
- [ ] Enough results for a rank in a category
- [ ] Holds across two snapshots (crowns only)

## What the lab claims

Self-reported. Shown for context; not used in any verdict.

- Lab claim: on ExploitGym, GPT-6 Cyber scores 48.5%, up from 32.1% for GPT-5.6 Cyber (about 51% relative improvement).
- Lab claim: in an alignment evaluation it avoided exploiting out-of-scope services in 99.93% of tests without its deployment prompt, and 100% with that prompt (up from 94.49% for GPT-5.6 Cyber without the prompt).
- Lab claim: used to develop 948 privilege-escalation exploits in Linux and to identify 81 previously unknown high-severity vulnerabilities in Firefox; on RemoteExploitBench it completed 95% of tasks while GPT-6 Astra with Daybreak Blue refused 100%.

## Sources

- LAB: [GPT-6 Cyber blog (OpenAI)](https://openai.com/index/introducing-gpt-6-cyber/) (2026-11-05)
- LAB: [OpenAI on X: GPT-6 Cyber](https://x.com/OpenAI/status/2118390190246879554) (2026-11-05)

Tracking ends Feb 3, 2027 unless the API or weights become available.

## Related

- Lab: [OpenAI](https://isitsota.com/labs/openai.md)
- Family: [GPT-6 family](https://isitsota.com/family/gpt-6.md)
- All pending releases: https://isitsota.com/releases.md
- JSON: https://api.isitsota.com/v1/models/gpt-6-cyber
