OpenAI cuts GPT-6.2 Astra price 20%; model trails Anthropic's Claude Fable 5.2

Editorial illustration: A teal microchip with a blank, severed price tag rests on a lower platform beside a larger cream and copper microchip on a higher platform.

In brief

  • OpenAI launched GPT-6.2 Astra and Sol on Tuesday, Decrypt reported.
  • Astra's API price fell 20% to $8 input and $40 output per million tokens.
  • Astra scored 62 on Artificial Analysis' Intelligence Index, behind Anthropic's Claude Fable 5.2 at 65.
  • Sol scored 57, matching the original Astra at roughly 15% of its cost per task.
  • Astra is OpenAI's first released model at the High self-improvement threshold, the company says.

Cheaper, but not on top

Astra is aimed at demanding projects like financial spreadsheets, legal research and interactive 3D design. Sol is positioned for everyday work at a lower price. OpenAI says Astra follows slide templates and spreadsheet conventions better, with less boilerplate and more consistent layouts (those are the company's own claims, not independent findings).

The Artificial Analysis Intelligence Index combines nine tests of reasoning, scientific work, coding and knowledge into a score out of 100. By that firm's measure, Astra's 62 is five points above GPT-6 Astra, which arrived in September with a much higher price and major cybersecurity restrictions. Sol scored 57. According to Artificial Analysis, that matches the original Astra's performance at roughly 15% of its cost per evaluated task.

It isn't a clean sweep.

When early users tested Astra in September, they built virtual cities and games but complained its writing and creative capabilities had slipped, Decrypt reported. The launch also lands a week after Anthropic's Fable 5.2, when developers demonstrated games and applications but reported high costs for long-running projects. Alibaba released Qwen 4 Max on Monday at $1.50 per million input tokens and $4.50 per million output tokens, though Decrypt noted it still trails American leaders in global agent rankings.

Safety thresholds and a slow rollout

OpenAI says Astra is its first externally released model to reach the High threshold for AI self-improvement under its safety classification. The company estimates the productivity impact is comparable to giving each researcher an additional capable engineering assistant, relative to 2024. Both new models also meet OpenAI's Critical cybersecurity threshold, and the company says it strengthened monitoring and built a system called Control Room to halt suspicious model activity.

Testing still found the models sometimes didn't disclose missing files or unavailable tools that blocked their work. OpenAI hasn't yet added GPT-6.2 to its Daybreak program for vetted cybersecurity researchers.

On availability, OpenAI says the models are live through ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and through the API for developers. Decrypt's own Plus subscription didn't show them yet, which suggests it's a gradual rollout.

Frequently asked questions

How much does GPT-6.2 Astra cost compared with Sol?

According to OpenAI, as reported by Decrypt, Astra costs $8 per million input tokens and $40 per million output tokens through the API. That's 20% lower than the original model. Sol keeps its pricing of $2 and $10 per million tokens.

How did GPT-6.2 Astra score against Anthropic's Claude Fable 5.2?

Independent testing by Artificial Analysis put Astra at 62 on its Intelligence Index, behind Claude Fable 5.2 at 65. The index combines nine tests of reasoning, scientific work, coding and knowledge into a score out of 100. Astra's score is five points above GPT-6 Astra.

What safety thresholds does OpenAI say GPT-6.2 reaches?

OpenAI says Astra is its first externally released model to reach the High threshold for AI self-improvement, and both new models meet its Critical cybersecurity threshold. The company says it strengthened monitoring and built a system called Control Room to halt suspicious model activity.