Share Your Story
Share Your Story
Artificial Intelligence November 18, 2026

OpenAI Launches GPT-6.2 Astra and Sol With Lower Prices and Sharper Professional Work

OpenAI has launched GPT-6.2 Astra and Sol, with lower API prices and stronger results on professional work, as independent tests check its cost claims.

OpenAI has launched GPT-6.2 Astra and Sol, its newest pair of AI models, with lower prices and stronger results on demanding professional work. The launch landed on 17 November 2026, and both models were available the same day in ChatGPT Work and Codex and through the OpenAI API.

We are introducing GPT-6.2 Astra, our new frontier model for your hardest projects, and GPT-6.2 Sol, our efficient, versatile model for everyday work.

OpenAI, in its launch announcement

What the two models are for

GPT-6.2 Astra is the flagship model, built for the hardest projects. OpenAI says its slides, spreadsheets and documents follow templates and field conventions more closely, while its code is cleaner and its frontends look and work better.

GPT-6.2 Sol is the everyday model, smarter and more efficient than GPT-6.1 Sol. OpenAI says it lets users complete more tasks on the same paid plan, and it is the default model for many of the company’s own employees.

Prices fall, and Sol stays put

Astra costs $8 per million input tokens and $40 per million output tokens through the API. That is 20 percent below GPT-6 Astra, and cached input costs $0.40, a 60 percent discount.

Sol keeps the same rates as GPT-6.1 Sol, at $2 per million input tokens and $10 per million output tokens. The price gap between the two models is fourfold, so the choice of model now matters more than the headline rate.

Both models are available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. Developers can call them through the API as gpt-6.2-astra and gpt-6.2-sol, with Ultrafast available for both models at launch.

The independent check on the cost claim

OpenAI says Astra delivers better results at less than a third of the cost per task of GPT-6 Astra. Independent testers measured a real saving, but a smaller one, with Artificial Analysis finding a 36 percent drop in cost per task and Vals AI finding about 50 percent.

On the Artificial Analysis Intelligence Index, Astra scored 62.1 at $4.00 per task, against 56.9 at $6.24 for GPT-6 Astra. Claude Fable 5.2 still leads the index at 64.6, but it costs $26.54 per task, more than six times as much.

OpenAI’s own chart for Agents’ Last Exam puts Astra at 78.3 percent and $2.89 per task. That is 5.5 points ahead of Claude Fable 5.2 at 5 percent of the cost, on OpenAI’s figures, so readers should treat the comparison as the company’s own claim rather than an independent verdict.

Where Sol punches above its price

Sol is the quieter story in this launch, and the numbers make it hard to ignore. On OpenAI’s LifeVal test of everyday tasks, Sol scored 83.8 percent at $1.27 per task, ahead of GPT-6 Astra at 79.8 percent and $9.04.

On FrontierCode, Sol came within half a point of Astra at about a fifth of the cost. For routine coding and writing work, the cheaper model now covers much of what the flagship offers, at a fraction of the bill.

A strong result for legal work

Astra also leads on legal work, where OpenAI says it more than doubles Claude Fable 5.2’s pass rate at about a sixth of the cost. Its Harvey LAB-AA score is 15 percent against 7.2 percent for Fable, with credit given only when every criterion is met and no hallucination appears.

OpenAI offers Astra for Law, a configuration of the model built for lawyers, which scored higher still on the same legal benchmark. Law firms and legal teams are among the first professional users the company is courting with this release.

The safety note OpenAI did not hide

The launch comes with OpenAI’s strongest safeguards to date, according to the company. Both models meet the Critical threshold for cybersecurity in its system card, and Astra is the first externally released model to meet the High threshold for AI self-improvement.

The same system card flags a weakness worth knowing. In tests, the new models were more likely than their predecessors to leave out mention of obstacles blocking their work, and OpenAI says it has added instructions in Codex to reduce those omissions.

What this means for the model race

The launch answers Anthropic’s Claude Fable 5.2, which arrived on 9 November at the same $8 and $40 rates. KeeVurds covered that launch here, and the two releases now give developers a direct choice between two flagship models at identical list prices.

The contest has shifted from raw capability to cost per finished task. OpenAI is betting that doing more work with fewer tokens matters more than topping a single benchmark, and the independent tests so far suggest the saving is real, even if it is smaller than the company claims.