Skip to main content

Kaer NewsSoftware

OpenAI releases GPT-6.2 Astra and Sol, cutting costs by a third

OpenAI launched GPT-6.2 Astra and Sol on 17 November, with Astra costing less than a third of GPT-6 Astra per task.

4 min read
OpenAI releases GPT-6.2 Astra and Sol, cutting costs by a thirdPhoto: 9to5Mac

Key points

OpenAI released GPT-6.2 Astra and Sol, with Astra costing less than a third of GPT-6 Astra per task and hitting a High self-improvement threshold.

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on 17 November, bringing both models to ChatGPT Work, Codex, and the API. Astra is the flagship for difficult projects, while Sol is the cost-effective option for everyday coding and writing. OpenAI says both can finish demanding tasks with fewer output tokens, letting users do more before hitting usage limits.

Astra costs less than a third of GPT-6 Astra per task while setting new highs on evals across professional work and scientific research, according to OpenAI. On Agents' Last Exam, which tests multi-stage professional tasks, Astra outperformed GPT-6 Astra by 5.2 points at less than one-third the cost. It also beat Anthropic's Claude Fable 5.2 by 5.5 points at just 5% of the cost, though the official announcement notes that Fable 5.2's leaderboard results exclude 9 tasks that used alternative models.

What the models offer

Sol is positioned as a reliable model for daily work, smarter and more efficient than GPT-6.1 Sol. On Cognition's FrontierCode 2.0, which measures whether code changes are good enough for maintainers to merge, Sol came within half a point of Astra at about one-fifth the cost. OpenAI says many employees use Sol as their go-to model for both technical and non-technical work.

Astra is tuned for high-quality professional output. OpenAI says its slides, spreadsheets, and documents read less like AI and more closely follow the templates and conventions users provide. It writes cleaner code and builds frontends that look and work better, particularly for websites, gaming, and 3D design. Early testers reported cleaner code style and more focused tests, with fewer rounds of revision, according to OpenAI.

GPT-6.2 Astra is the first externally deployed model to meet the High threshold for AI self-improvement under OpenAI's Preparedness Framework, the company said. That framework grades dangerous capabilities against tiers, and self-improvement asks whether a model can contribute to the development of AI itself, potentially speeding up its own successors. GPT-6 Astra, released in September, fell short of that threshold despite a perfect score on ExploitBench.

Safety classifications

Sol did not reach High on self-improvement. Instead, it meets the Critical threshold for cybersecurity and the High threshold for biological and chemical risks, according to OpenAI's classifications. These labels come from OpenAI's own framework, making them a self-reported warning label rather than an outside audit. Both models are available to all Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, with Enterprise and Edu requiring admin enablement. The $200 per month Pro plan gets Ultrafast mode for the first time with Sol, while the $500 per month Pro plan gets Ultrafast on both models. Ultrafast can generate tokens at up to 4.5 times the standard rate, though it uses more of the user's allowance.

Pricing and availability

Developers can use both models through the API today. Astra's standard input and output prices are 20% lower than with GPT-6 Astra, with pricing at $8 per million tokens for input, $0.40 for cached input, and $40 for output. Sol keeps its GPT-6.1 pricing at $2 for input, $0.10 for cached input, and $10 for output. Both models support a long context of 1.05 million tokens, but inputs exceeding 272,000 tokens double the input price and increase output by 1.5 times.

OpenAI is also cutting Ultrafast pricing by 50% for the next three months for API and Enterprise users, making it available at 3 times the cost of Standard. That 3x rate is valid until 28 February 2027, after which it returns to 6x starting 1 March. The release comes a week after Anthropic launched Claude Fable and Mythos 5.2, which the company said beat GPT-6 Astra virtually across the board in benchmarks.

OpenAI says GPT-6.2 Astra and Sol are more or less playing ball with Anthropic's latest models, often delivering similar performance at a drastically lower cost. The company recommends Astra for the hardest projects, while Sol is the go-to for everyday work. Both models are now live, and users can switch speeds from the model picker.

Frequently asked questions

What are GPT-6.2 Astra and Sol?

GPT-6.2 Astra is OpenAI's flagship model for difficult projects, while Sol is a cost-effective model for everyday coding and writing. Both were released on 17 November 2026 and are available in ChatGPT Work, Codex, and the API.

How much does GPT-6.2 Astra cost?

GPT-6.2 Astra costs $8 per million tokens for input, $0.40 for cached input, and $40 for output. This is 20% lower than GPT-6 Astra's pricing, and the cost per task is less than a third of GPT-6 Astra.

What is the self-improvement threshold for GPT-6.2 Astra?

GPT-6.2 Astra is the first externally deployed model to meet the High threshold for AI self-improvement under OpenAI's Preparedness Framework. This means it can contribute to the development of AI itself, potentially speeding up its own successors.

More from this edition