Where the models are available
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on 17 November. The company lists both in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and in its developer API. It says eligible Enterprise and Edu users get access once their workspace administrator enables it.
The standard API prices per million input and output tokens are $8 and $40 for Astra, and $2 and $10 for Sol. A token is a small unit of text or other model input or output. Sol keeps the rates of GPT-6.1 Sol; Astra’s regular input and output rates are 20% lower than GPT-6 Astra’s. A lower rate does not guarantee a lower bill on every task: the total depends in part on how much the model reads and writes.
What the cybersecurity rating means
OpenAI says both models were better than their predecessors in its tests at turning known software vulnerabilities into working exploits. It puts both in the Critical category for cybersecurity under its own safety framework, a threshold the earlier GPT-6 Astra had already reached. The rating describes what the models can do under test conditions; it is not an independent finding about how often harm occurs.
The company says both models have tighter default limits on higher-risk cybersecurity assistance. Its system card says the more permissive configuration it plans for GPT-6.2 Astra is limited at launch to organisations separately approved for Daybreak Red, a programme for authorised security work. Daybreak Blue approval alone does not lower the safeguards on these models.
What the tests leave open
Several published cyber scores came from tests with production safeguards turned off. Those results can help explain why OpenAI tightened access, but they do not describe what a standard account will answer. OpenAI acknowledges that the stricter rules can also limit legitimate security research and testing. The documents do not show how often that trade-off will affect real work.
Sources
- OpenAI: Introducing GPT-6.2 Astra and SolRelease and named access channels, standard API prices and what the internal cyber tests disabled.
- OpenAI: GPT-6.2 system cardCapability classifications, changes to default access, Daybreak vetting and the limits of the company’s tests.



