Model launches

AI-generated text

OpenAI introduces GPT-6.2 Astra and Sol models

OpenAI has released GPT-6.2 Astra for demanding projects and GPT-6.2 Sol for everyday work through its API.

OpenAI introduces GPT-6.2 Astra and Sol models

OpenAI has made two new models available through its API: GPT-6.2 Astra is a frontier model intended for the most demanding projects, while GPT-6.2 Sol is an efficient, versatile model for everyday work. They are available under the identifiers gpt-6.2-astra and gpt-6.2-sol; version 0.38 of the LLM tool also supports them.

Pricing

GPT-6.2 Astra costs $8 per million input tokens and $40 per million output tokens. That is 80% of the price of GPT-6 Astra and matches the pricing of Claude Fable 5.2. GPT-6 Astra and Claude Fable 5 each cost $10 per million input tokens and $50 per million output tokens.

The price of cached input has fallen by a larger margin, from $1 to $0.40 per million tokens. That could be particularly useful for longer agentic sessions, in which models work autonomously on tasks.

GPT-6.2 Sol costs $2 per million input tokens and $10 per million output tokens, the same as GPT-6.1 Sol and GPT-6 Sol. All three are half the price of GPT-5.6 Sol, which costs $4 and $20 respectively.

Tests and early feedback

The source article's author also tested both models on an animation task in Codex CLI, using the “Ultra” effort level. The prompt requested a Three.js animation in which Yonkers transforms into a giant mech and stomps through Manhattan, Queens and Brooklyn, using OpenStreetMap mapping data. The author found that neither model transformed Yonkers convincingly enough, particularly compared with results from Claude Fable 5.2 and Gemini 4 Carbon.

To show the two animations simultaneously, the author also used GPT-6.2 Astra in the Codex desktop app. The request was for a square, 2,000-by-2,000-pixel video with the Astra animation on top, the Sol animation below and labels for each. The initially requested duration of ten seconds was later changed to twenty seconds.

Early reception has been mixed. According to the source, Astra 6.2 scores below Fable 5.2 on several benchmarks, and users have also complained about its communication style. In feedback sent to OpenAI before release, Theo Browne said the model can manage multiple subagents well during long tasks but sometimes asks users about internal implementation details on which they cannot reasonably provide input. As an example, he described a request to create an Outer Wilds level: after two hours, the work stalled because the model asked him to choose between two internal filenames.