OpenAI has released GPT-6.2 Astra and GPT-6.2 Sol, the latest refresh of its flagship and everyday frontier models, bringing both to ChatGPT and Codex starting today. The launch, confirmed by Bloomberg and covered by The Verge, arrives weeks after the company canceled a prior release of the update, making GPT-6.2 one of the most closely watched model rollouts since the GPT-6 generation began with Astra in September.

The announcement follows weeks of speculation about when the update would ship, and it lands as the biggest AI model news event of the day for developers and ChatGPT subscribers alike.

What's New in GPT-6.2 Astra

OpenAI continues to position Astra as the model for the hardest workloads. In the announcement highlighted by 9to5Mac, the company demonstrated using a screenshot of a slide as a template, with GPT-6.2 Astra matching the layout, colors, and even the font size for the rest of a presentation. On the coding side, OpenAI says Astra is better at making and testing changes within an existing project, and that early testers reported "cleaner code style and more focused tests, with fewer rounds of revision."

Both models are also designed to finish demanding tasks with fewer output tokens, which OpenAI says lets users get more done before hitting their usage limits.

Sol Closes the Gap at a Fraction of the Price

Sol remains OpenAI's more affordable option for everyday coding and writing, but the gap to Astra is narrowing. OpenAI says that on Cognition's FrontierCode 2.0 benchmark — which measures whether changes to real codebases are good enough for maintainers to merge — GPT-6.2 Sol came within half a point of Astra at maximum reasoning effort, at about one-fifth of the cost.

For teams weighing frontier quality against budget, that trade-off is becoming harder to argue with.

Availability, Plans, and Ultrafast Mode

GPT-6.2 Astra and Sol are available starting today in ChatGPT and Codex for Plus, Pro, Enterprise, and Edu customers. The rollout includes a new performance tier:

  • $200/month Pro plan: gains Ultrafast mode with Sol for the first time
  • $500/month Pro plan: gets Ultrafast on both models
  • Enterprise and Edu: Ultrafast becomes available once an workspace admin enables it

OpenAI says Ultrafast generates tokens at up to 4.5 times the standard rate on either model, though it consumes more of a user's allowance. Speeds can be switched from the model picker.

On the API side, OpenAI has cut Astra's standard input and output prices by 20% compared with GPT-6 Astra, while Sol keeps its GPT-6.1 pricing. The GPT-6.2 Astra listing on OpenRouter shows pricing of $8 per million input tokens and $40 per million output tokens.

A System Card Heavy on Cyber Safeguards

Alongside the launch, OpenAI published a full system card for the GPT-6.2 models on its Deployment Safety Hub, and the document shows how much safety machinery now surrounds Astra-class releases. The preparedness assessment includes evaluations across High biological capabilities, Critical cybersecurity capabilities, and High AI self-improvement capabilities — the top tier of OpenAI's tracking framework.

The system card describes external evaluations by the UK AI Safety Institute on cyber, alignment, and monitorability, biological safeguard assessments with SecureBio, and restricted "trusted access" programs for the most sensitive capabilities, including a dedicated biology program under the GPT-Rosalind track and trusted access controls for cybersecurity work. Model-level mitigations include misalignment monitoring, distillation mitigations, jailbreak red-teaming, and safe completions designed to refuse harmful tasks without degrading benign ones.

OpenAI has treated Astra-class models with unusual caution since the summer, when the company slowed Astra's release as it neared a first-ever "critical" cyber capability threshold and later restricted access to the original GPT-6 Astra over its advanced cyber capabilities. The depth of the GPT-6.2 documentation suggests that approach has hardened rather than eased.

A Launch That Almost Wasn't

The rollout closes out a rocky pre-release period. Bloomberg reported that the new Astra model arrives after OpenAI scrapped a prior release, and the update notably failed to ship on its mid-November expected window, with industry outlets documenting that GPT-6.2 had not launched alongside the announcements OpenAI made on November 12. The GPT-6 generation began with Astra in September, followed by the Sol line's upgrade to GPT-6.1 at the company's DevDay a few weeks later.

With GPT-6.2 finally in users' hands, attention now shifts to how the models perform in real-world coding, agentic, and everyday workloads — and whether the scrapped first attempt made the shipped product measurably safer.

Astra or Sol: How to Choose

For subscribers deciding where to spend their usage, the division of labor between the two models is clearer than it was in the GPT-6.1 era. Astra remains the recommendation for the hardest projects — long agentic sessions, subtle refactors across a live codebase, and multimodal work like rebuilding a deck from a single reference screenshot. Sol is positioned for the high-volume, lower-stakes work: everyday coding assistance, drafting, and iteration where near-Astra quality at a fifth of the cost changes the economics entirely.

The FrontierCode 2.0 result is the number most developers will watch. A gap of half a point, at maximum reasoning effort, on a benchmark that measures merge-worthy changes to real repositories, suggests that for many production tasks Sol can now do most of what Astra does — and users of OpenAI's Codex coding agent can route accordingly. OpenAI's own framing of the launch leaned into that efficiency story: "GPT-6.2 Astra and Sol are now available. Super efficient and built to perform where it counts—the real world," the company posted on X alongside the announcement.

Ultrafast mode adds a third lever. At up to 4.5 times the standard token generation rate, it targets the interactive use cases where latency matters more than throughput — though users should expect it to draw down their allowance proportionally faster.

Stay Ahead of AI

Track every frontier model release as it happens: Read more AI news →