OpenAI expanded its Daybreak program to include two separate access tiers, offering frontier general-purpose models through one channel and specialized vulnerability research capabilities through another, according to the announcement posted today.
Daybreak Blue grants approved individuals and organizations access to GPT-5.6 Sol, the company's frontier general-purpose model, for authorized work. Daybreak Red provides access to GPT-5.6-Cyber, a new model designed specifically for authorized vulnerability research, exploit validation and security testing. Both tiers restrict access to pre-approved parties only.
The program creates formal separation of frontier model access into two use cases: unrestricted general-purpose capability through Blue, and controlled security research through Red. The structure mirrors OpenAI's earlier approach to managing high-capability model distribution, but the introduction of a model specifically trained for vulnerability discovery differs from treating all frontier access identically.
GPT-5.6-Cyber is purpose-built to assist security researchers and defenders in identifying vulnerabilities before threat actors can exploit them. OpenAI designed the model with constraints intended to prevent misuse while enabling legitimate defensive work. The company has not disclosed technical details of those constraints or quantitative restrictions on query types.
Daybreak Blue includes GPT-5.6 Sol beyond the closed research setting where earlier versions were tested. The program does not appear to include a public tier; both access levels require organizational vetting and demonstration of authorized use.
OpenAI maintains direct control over approval decisions for both tiers. The company did not specify approval timelines, review criteria, or rejection rates. No information was provided on the number of organizations currently enrolled in either tier or whether enrollment is capped.
