enzosoler.comreviews 09bench 9 deviceslive test Can a ring and a watch replace my Apple Watch?
Published

GPT-6.2 Astra cuts OpenAI's top API price 20%, and Sol nearly matches it on code

A laptop on a wooden desk in a dim room, showing lines of code in an editor
Image: Daniil Komov / Pexels

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17 in ChatGPT Work, Codex and the API. Astra costs $8 per million input tokens and $40 per million output, 20% below GPT-6 Astra. Sol stays at $2 and $10. Plus, Pro, Enterprise and Edu get both today, and Pro adds an Ultrafast mode.

If you call OpenAI’s top model through the API, switch to GPT-6.2 Astra and the same job costs 20% less. OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on Tuesday, November 17, in ChatGPT Work, Codex and the API, according to OpenAI’s announcement. Before you pay for Astra at all, test Sol: OpenAI’s own coding result puts it within half a point of Astra for about a fifth of the cost.

What changed

  • Two new models. Astra is the model OpenAI recommends for the hardest work: documents, slides and changes to existing code. Sol is the cheaper model for everyday coding and writing, as 9to5Mac reported.
  • Fewer tokens per task. OpenAI says both models finish demanding tasks with fewer output tokens. That cuts cost on top of the price change.
  • Slides from a screenshot. In ChatGPT, Astra can take a screenshot of a slide as a template and match its layout, colors and font size, according to 9to5Mac.
  • Codex. OpenAI says Astra is better at making and testing changes in an existing project. Early testers reported cleaner code and more focused tests with fewer rounds of revision.

What it costs on the API

Prices per million tokens, as 9to5Google reported on November 17:

Model Input Cached input Output
GPT-6.2 Astra $8 $0.40 $40
GPT-6 Astra $10 n/a $50
GPT-6.2 Sol $2 $0.10 $10

Sol keeps the price of GPT-6.1 Sol. A job that reads 200,000 tokens and writes 20,000 costs $2.40 on GPT-6.2 Astra, $3.00 on GPT-6 Astra and $0.60 on GPT-6.2 Sol.

These are the same list prices Anthropic set a week earlier. Claude Fable 5.2 costs $8 and $40, and Sonnet 5.5 costs $2 and $10.

Who gets it in ChatGPT

Plus, Pro, Enterprise and Edu users get both models starting today in ChatGPT Work and Codex, according to 9to5Mac and 9to5Google. Neither report names the Free or Business plans.

Ultrafast mode generates tokens up to 4.5 times faster on either model and uses more of your allowance. You pick it in the model picker:

  • Pro at $200 a month: Ultrafast on Sol only.
  • Pro at $500 a month: Ultrafast on both models.
  • Enterprise and Edu: both models, once an admin turns it on.

If you are watching usage limits, OpenAI doubled the Work and Codex allowances on the Pro plans earlier this month (our report). Ultrafast spends that allowance faster.

Sol is the default to beat

On Cognition’s FrontierCode 2.0, which checks whether a change to a real codebase is good enough for maintainers to merge, Sol landed within half a point of Astra at about one fifth the cost, both at maximum reasoning effort, according to 9to5Mac. OpenAI has not published the raw scores, so treat that as the company’s claim until they are out.

Our read: for a small team, the practical move is to run Sol by default and send only the jobs that fail on Sol up to Astra. The price gap is four times per token. If Sol merges cleanly for your codebase, Astra’s quality edge costs more than it returns.

Three things to do this week:

  1. Change the model name in your API calls if you use GPT-6 Astra. The 20% cut only applies to the new model.
  2. Rerun one real task on Sol before you renew a Pro plan for Astra access. Plus already includes both.
  3. Compare against your Claude bill. With Astra and Fable 5.2 at the same list price, the deciding number is tokens per finished task on your own work. Run the same job on each and compare the invoices.

OpenAI’s previous price move, in October, halved the cost of its mid-size model (GPT-6.1 Terra). Google’s Gemini 4 Hydrogen sits below all of these at $1.50 per million input tokens.

One email when a review goes up. Unsubscribe from any of them.