OpenAI’s GPT-6.2 Astra Is Here: Smarter for Less, and GPT-6.2 Sol Costs $2 per Million Tokens
OpenAI just shipped two new models, GPT-6.2 Astra and GPT-6.2 Sol. The pitch: frontier-level results for a lot less money. OpenAI says Astra beats GPT-6 Astra on professional and science work “at less than a third of the cost per task,” and Sol nearly matches Astra on real coding work at about one-fifth the cost.
Where you can use them
According to OpenAI’s Nov. 17, 2026 announcement, both models are live in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and in the API as gpt-6.2-astra and gpt-6.2-sol. Pro 500 subscribers get the faster Ultrafast mode on both, and Pro 200 gets Ultrafast starting with Sol.
One catch: an OpenAI forum moderator said the rollout “does not add either model to regular Chat,” and availability depends on your plan and rollout progress. If you don’t see it, update the desktop app or the Codex CLI.
The prices
| API model | Input (per 1M tokens) | Output (per 1M tokens) | Best for |
|---|---|---|---|
| GPT-6.2 Astra | $8.00 | $40.00 | Hardest projects, science, research |
| GPT-6.2 Sol | $2.00 | $10.00 | Everyday work, coding |
OpenAI says Astra’s price is 20% lower than its predecessor’s, per The Decoder. Sol keeps GPT-6.1 Sol’s price.
What OpenAI claims
- Agents’ Last Exam: Astra scores 5.2 points above GPT-6 Astra at less than half the cost per task.
- FrontierCode 2.0: Sol “nearly matches Astra on merge-ready changes to real codebases at about one-fifth the cost.”
- Cybersecurity: Astra is “our most capable model ever for cyber,” with a 22-point jump on an internal exploit benchmark.
What independent testers found
Outside benchmarks, collected by The Decoder, are more mixed:
- Artificial Analysis puts Astra second on its Intelligence Index (62), behind Claude Fable 5.2 (65), and says OpenAI holds the best cost-for-performance spot. But coding is flat: Astra matches GPT-6 Astra on its Coding Agent Index, and Sol drops 3 points from GPT-6.1 Sol.
- Vals AI measured about $10.68 per task for Astra vs. $21.28 for GPT-6 Astra, and ranked Astra first on its science index, but low on legal research.
- Epoch AI scored Astra level with Fable 5.2 on its capability index, but said it barely improved on GPT-6 Astra in real-world project tasks. It also solved 7 of 68 FrontierMath Erdős problems, a new record.
The safety fine print
OpenAI rates both models at the “Critical” level for cybersecurity under its own Preparedness Framework, the same as GPT-6 Astra, so they launch with tighter safeguards on risky security requests. That can also slow legitimate security research; OpenAI says vetted defenders will get access with fewer refusals through its Daybreak program. Astra is also the first model OpenAI has released at its “High” level for AI self-improvement, meaning OpenAI says it can meaningfully speed up AI research. These are OpenAI’s own ratings, per The Decoder’s read of the system card.
PCGeek’s take
The story here is price. If Sol really gets close to Astra on everyday coding for a fifth of the cost, that’s the model most people should try first. Just keep the hype in check: most of the big numbers are OpenAI’s own, and independent testers say coding didn’t move much this round. Try it on your own work before you switch.
More on OpenAI this week: California sues OpenAI over the Hugging Face hack.
Will you try Astra or stick with the cheaper Sol? Tell us in the comments.
Follow @pcgeekinfo for more.
Featured image: “Person holding the OpenAI icon” by FoxTPNL, Wikimedia Commons, CC BY 4.0. File photo.