OpenAI Launches GPT 6.2 Astra and Sol, Its New Frontier Pair
Some of you have been asking very politely. Some of you have just been asking. GPT-6.2 Astra and Sol are now available. Super efficient and built to perform where it counts—the real world.
Context
A new official frontier pair is live in ChatGPT, Codex, and the API, and early testers immediately split over quality and price.
OpenAI posted that GPT 6.2 Astra and Sol are now available and linked to a company blog. Follow up posts describe Astra as a frontier model for hard projects and Sol as an efficient everyday model. OpenAI Developers said both are in Codex and the API. Testers Matt Shumer and Theo said their early access went badly, which is attributed opinion, not an official verdict.
- OpenAI announcement post
- OpenAI blog introducing GPT 6.2 Astra and Sol
- OpenAI follow up on GPT 6.2 Astra
- OpenAI follow up on GPT 6.2 Sol
- OpenAI Developers post on Codex and the API
- Matt Shumer's tester post
- Theo's tester post calling Astra the worst OpenAI model
- Tibo's post on Sol and ultrafast for Pro 200 plans
Deep dive
OpenAI announced that GPT 6.2 Astra and Sol are now available, calling them super efficient and built to perform in the real world, and linked to a company blog. A follow up post said Astra is a frontier model for the hardest projects, sets new highs on evaluations of professional work and scientific research, and costs less than a third per task compared with GPT 6 Astra. A third post said Sol is an efficient everyday model that handles multistep agent work, and that many OpenAI employees use it as their go to model.
The company blog says Astra scored 5.2 points higher than GPT 6 Astra on Agents Last Exam at less than a third of the cost, and 5.5 points ahead of Claude Fable 5.2 at 5 percent of the cost per task. It also reports a new high on Terminal Bench Science 0.1 and on the Vals Science Index, and says Astra's API input and output prices are 20 percent lower than GPT 6 Astra, with a 60 percent discount on cached input. Sol, the blog says, keeps GPT 6.1 Sol prices and comes within 0.5 points of Astra on Cognition's FrontierCode 2.0 at about one fifth the cost.
OpenAI Developers said both models are now in Codex and the API. Tibo, who leads Codex, wrote that Sol and Astra are finally here and that ultrafast was expanded to Pro 200 plans for Sol. Those are official or staff posts.
The tester reaction is a different matter. Matt Shumer said he would not write a full review because his experience testing the models was terrible and they were essentially unusable. Theo wrote that he had been testing Astra for a while and thought it was the worst model OpenAI has ever released, and later described two testers independently getting similar 'Bellwether' game levels. Those comments are attributed opinions from early testers, not a company finding.
What to expect next
Usage, price, and independent benchmarks over the next few weeks will show whether Astra and Sol land closer to OpenAI's blog or to the tester backlash. Watch pricing in the API, how often Sol becomes the default at work, and whether later patches change the early quality complaints. Rivals at Anthropic, Google, and Microsoft will be measured against the new pair, and Codex customers will see whether the Pro 200 ultrafast change actually improves daily agent work.
