OpenAI Unveils GPT-6.2 Astra and Sol, Outperforming Claude Fable 5.2 at Just 5% of the Cost

11/18 2026 560

Kuai Technology, November 18 — OpenAI has released two new models: GPT-6.2 Astra and GPT-6.2 Sol. Astra is positioned as a frontier model for highly demanding projects, while Sol is designed as an efficient general-purpose model for everyday work.

Both models deliver stronger performance on highly difficult tasks while using fewer output tokens, offering scalable frontier-level capability for enterprises and developers.

On the Agents' Last Exam, which evaluates multi-step professional work, Astra scores 5.2 points higher than GPT-6 Astra at less than one-third the per-task cost, and 5.5 points higher than Anthropic's Claude Fable 5.2 at only 5% of the cost.

On Terminal-Bench-Science 0.1, which assesses scientific research capabilities, it outperforms the previous generation by 9 percentage points, also at less than one-third of the cost.

In terms of product capabilities, slides, spreadsheets, and documents generated by Astra better match user templates and industry standards; code style is cleaner; and performance is stronger in frontend, website, game, and 3D design tasks.

Computer use can reach deep into the component level of CAD and simulation tools, and both models can turn text briefs into interactive 3D experiences in the browser.

Regarding safety thresholds, both models' cybersecurity capabilities reach the Critical threshold of the Preparedness Framework, and their biological and chemical capabilities reach the High threshold.

Astra is OpenAI's first publicly released model whose AI self-improvement capability reaches the High threshold. The company has implemented supporting safeguards, including research environment hardening, Control Room workload tracking, and expanded monitoring. In alignment evaluations, the two models perform comparably to GPT-6 Astra and exhibit fewer inappropriate behaviors in simulated workplace tasks.

In terms of pricing, for the API, Astra costs $8 per million input tokens (about 57 RMB), $0.40 for cached input (about 2.9 RMB), and $40 for output (about 286 RMB). Input and output prices are 20% lower than GPT-6 Astra, with a 60% discount on cached input.

Sol follows GPT-6.1 Sol pricing: $2 input (about 14.3 RMB), $0.10 cached input (about 0.7 RMB), and $10 output (about 71.5 RMB).

Both models are available starting today to Plus, Pro, Enterprise, and Edu users of ChatGPT Work and Codex, and are simultaneously available via API. They have not yet entered the regular chat mode.

Solemnly declare: the copyright of this article belongs to the original author. The reprinted article is only for the purpose of spreading more information. If the author's information is marked incorrectly, please contact us immediately to modify or delete it. Thank you.