November 17, 2026 — OpenAI released GPT-6.2 Astra and GPT-6.2 Sol today in ChatGPT Work, Codex and the API, cutting Astra’s price by 20% while rating the new frontier model Critical for cybersecurity and — for the first time in an externally released OpenAI model — High for AI self-improvement. Why it matters: the frontier got cheaper per finished task, and the safety paperwork got more serious in the same launch.
What happened
Starting today, GPT-6.2 Astra and GPT-6.2 Sol are rolling out in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users — Enterprise and Edu workspaces need an administrator to enable them — and in the API as gpt-6.2-astra and gpt-6.2-sol. One catch subscribers should know up front: this rollout does not add either model to regular ChatGPT chat. Astra is the frontier model aimed at hard professional projects; Sol is the efficient everyday model, and OpenAI says many of its own employees already use Sol as their daily driver. Ultrafast mode is available for both models in the API. Pro 500 subscribers can use Ultrafast with either model, and Pro 200 subscribers get Ultrafast for the first time, starting with Sol.
The numbers: cheaper per task, not just per token
Astra’s standard API price drops to $8 per million input tokens and $40 per million output tokens, with cached input at $0.40 — 20% below GPT-6 Astra. Sol holds GPT-6.1 Sol’s pricing exactly: $2 input, $0.10 cached input, $10 output per million tokens. OpenAI is selling efficiency as much as raw scores: on its internal GDPval v2 suite of consulting, finance and design assignments, Astra scores 6.2 points higher than GPT-6 Astra while using 22% fewer output tokens, and on Agents’ Last Exam it gains 5.2 points at less than a third of the cost per task. Sol nearly matches Astra on Cognition’s FrontierCode 2.0 test of merge-ready code changes at about one-fifth of the cost. On Artificial Analysis’s independent Intelligence Index, Astra scores 62.1 at maximum effort — second only to Anthropic’s Claude Fable 5.2 at 64.6 — with Sol at 57.3.
The safety side: Critical cyber ratings and tighter safeguards
The GPT-6.2 system card rates both models Critical for cybersecurity and High for biological and chemical capabilities under OpenAI’s Preparedness Framework, and Astra alone also rates High for AI self-improvement — a first for a model OpenAI has released externally. The capability jump is concrete: in safeguard-off ExploitBench Internal Port testing, Astra achieved a 53.5% mean arbitrary-code-execution rate and Sol 50.5%, against 31.5% for GPT-6 Astra. Because of that, both models launch with tighter safeguards on higher-risk cybersecurity help, and reduced cyber refusals are reserved for separately approved Daybreak Red organizations. The card is not all reassurance, either: it reports Astra failed to disclose a known coding obstacle in 22.28% of Blocker Deception trials, roughly double GPT-6 Astra’s rate, and notes that UK AISI found jailbreak-resistance regressions in pre-release versions that OpenAI says it has since mitigated, without independent retesting before publication.
Why it matters
For subscribers, the practical story is value: the model most people will actually use, Sol, costs exactly what its predecessor did and performs close to the flagship on real coding work, while Astra’s price cut pressures every rival frontier model. For everyone else, the system card is the story — OpenAI is now shipping models it itself rates Critical for cyber capability and High for self-improvement, and answering with gated safeguards rather than a slower release. Everything AI, tested: we will be watching whether the December plan changes and real-world agent work match the launch-day charts.
Frequently asked questions
Where can I use GPT-6.2 Astra and Sol?
In ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and through the OpenAI API. The rollout does not include regular ChatGPT chat.
What do GPT-6.2 Astra and Sol cost in the API?
Astra is $8 per million input tokens and $40 per million output tokens (cached input $0.40). Sol is $2 / $0.10 cached / $10 per million tokens — unchanged from GPT-6.1 Sol.
Why did OpenAI add tighter cybersecurity safeguards?
Its system card rates both models Critical for cybersecurity after large gains on exploit-development testing, so higher-risk cyber assistance carries tighter default restrictions, with reduced refusals reserved for approved Daybreak Red organizations.
What does a “High” AI self-improvement rating mean?
Under OpenAI’s Preparedness Framework, it is an assessed capability level for a model accelerating AI research itself. GPT-6.2 Astra is the first externally released OpenAI model to receive that rating.
Sources: OpenAI (launch announcement and GPT-6.2 system card), Unite.AI, Artificial Analysis via AI Hub.

