Daybreak (cyber defense program) and GPT-5.5-Cyber
Daybreak bundles trusted-access cyber models, Codex Security and partners to move defenders from vulnerability findings to validated patches.
Pairs GPT-5.5-Cyber (full version) with Codex Security workflows and an open-source "Patch the Planet" effort. Daybreak split into Blue (general models) and Red (purpose-trained GPT-5.6 Cyber for authorized exploit work) tiers on 2026-08-07; $1B for frontline defenders followed 2026-09-03.
- Date
- Monday, 22 June 2026
- Lab
- OpenAI
- Kind
- product
- Access
- research preview
- Price
- access by application; Rosalind/Daybreak pricing in the changelog
Figures
| Measure | Value | Measured by |
|---|---|---|
| Codex Security cloud scale since March | 30 million+ commits, 30,000+ codebases 70,000+ findings marked fixed by humans | company |
The cyber-capability trajectory (GPT-5.3-Codex first High, GPT-5.6 High, GPT-6 Astra Critical) is what drove the tiered-access model (see openai-gpt-6-astra).
Sources
- openai.com/index/daybreak-securing-the-world/
- developers.openai.com/api/docs/changelog
- openai.com/index/the-defenders-window/
This record was checked against its sources on 6 October 2026. How we check