OpenAI has released two new models, GPT-6.2 Astra and GPT-6.2 Sol. Astra is aimed at the hardest professional projects, while Sol is built for everyday work efficiency. Both are live in ChatGPT Work and Codex, and ChatGPT Plus, Pro, Enterprise and Education users can select them too. Developers can call gpt-6.2-astra and gpt-6.2-sol through the API, with both the Responses and Chat Completions endpoints supported.
Pricing: Astra drops 20%, Sol stays put
In the API, GPT-6.2 Astra costs $8 per million input tokens, $0.4 per million cached input tokens and $40 per million output tokens. Compared with GPT-6 Astra, input and output prices are each 20% lower, and cached input is down 60%.
GPT-6.2 Sol keeps the pricing of GPT-6.1 Sol: $2 per million input tokens, $0.1 per million cached input tokens and $10 per million output tokens. Side by side, all three of Sol's rates are exactly one-quarter of Astra's.
A rough calculation: a team that burns through 10 million input tokens and 2 million output tokens a day would pay about $160 a day on the new Astra, versus roughly $200 at GPT-6 Astra's old prices. If the workload can be handed to Sol, the bill falls to about $40 a day. This ignores cache hits, so the real invoice will depend on how much of the prompt gets reused.
Benchmarks published by OpenAI
According to OpenAI's launch page:
- On the internal GDPval v2 evaluation, GPT-6.2 Astra scores 6.2 percentage points higher than GPT-6 Astra and uses 22% fewer output tokens;
- On the same evaluation, GPT-6.2 Sol beats GPT-6.1 Sol by 5.3 points;
- On Cognition's FrontierCode 2.0, the gap between Sol and Astra is within 0.5 points, at roughly one-fifth of Astra's cost;
- In a set of Elo rankings shared by OpenAI, GPT-6.2 Astra ranks first at 1318, ahead of Claude Fable 5.2 at 1271 and GPT-6.2 Sol at 1259.
These are all OpenAI's own test results, and third-party leaderboards have not yet rerun them. GDPval v2 is an internal OpenAI evaluation, and outsiders cannot fully reproduce it.
Astra skipped 6.1
The naming shows that OpenAI's two product lines move at different paces. Sol went from GPT-6 Sol to GPT-6.1 Sol and now 6.2, with each version released publicly. Astra jumped straight from GPT-6 Astra to 6.2, after the GPT-6.1 Astra release originally planned for October was pulled.
Before GPT-6.2 Astra officially launched, some Codex users had already spotted a gpt-6.2-astra entry in their usage statistics, while the model picker still showed GPT-6 Astra. With this launch, 6.2 Sol and 6.2 Astra now sit at the top of the model picker, and older versions such as 6.1 Sol and GPT-6 Astra remain selectable.
For developers already running GPT-6 Astra, switching to 6.2 Astra means a lower unit price and shorter outputs for the same tasks. By OpenAI's account, Sol is already close to Astra on coding tasks, so more work may be a fit for Sol. Which tasks justify paying four times the rate will depend on what each team measures on its own workloads, and OpenAI's launch page offers no specific guidance on choosing between the two.
Sources: OpenAI launch page, RuntimeWire, CocoLoop, TestingCatalog. API prices follow OpenAI's published standard rates per million tokens; benchmark figures are as measured by OpenAI.