Notes on film, new tools and the decisions behind the work.

Tools / Analysis

OpenAI releases GPT-6.2 Astra and Sol. When is Astra worth the extra cost?

Both models are rolling out in ChatGPT Work and Codex. A retailer’s slide deck shows what Astra adds; a coding test shows how close the less expensive Sol can get.

A receipt splits into two ribbons. SOL curls into a loop beside the straight ASTRA strip above the GPT-6.2 headline.
DU / SIGNALS · S/117Signals illustration · Google Flow · GPT-6.2 release from OpenAI

OpenAI released GPT-6.2 in two versions on November 17. Astra is the model the company recommends for its hardest projects. Sol is the one it recommends for everyday work, at a quarter of Astra’s standard API price for the same number of tokens. Both are rolling out in ChatGPT Work and Codex, and developers can call either through the API.

One way to see what the premium buys is to look at a slide both generations of Astra got right. The difference was which number the newer model decided to put in the headline.

Two models read the same accounts

OpenAI gave the previous Astra and GPT-6.2 Astra a retailer’s first-quarter presentation, second-quarter results and regulatory filing. The assignment was to update the deck. Both models identified the refund of import duties as the main reason earnings rose. That matters because a refund can lift profit even if customers buy nothing more.

The newer Astra led with a different figure: sales grew 5.3%. It separated online growth from growth in stores and advised managers to build a budget that did not assume another refund. A reader could distinguish what happened at the checkout from what happened in the accounts. The launch page puts both decks side by side; it is a selected company example.

Sources: OpenAI [1] · OpenAI / X [4]

X / OPENAI

OpenAI announces both models

Loading post from X…

Credit / OpenAI / @OpenAIOpen original 

The company’s launch post links to the examples and evaluation results.

Where Sol comes close

On Cognition’s FrontierCode 2.0, which checks whether changes to existing software meet the requirements, OpenAI reports a gap of half a point or less between Sol and Astra. Both ran at maximum reasoning effort. Sol’s run cost about one-fifth as much. That figure is the cost observed on this test, not a new discount or a guarantee about the next repository.

For a team deciding which model should handle routine code changes, that makes Sol a useful first trial. For work that gets delivered as a document, the trial needs a different question. Give both models the same source files and template, then count the corrections a reviewer must make before sending the result. Those corrections have a cost that a token invoice will not show.

Sources: OpenAI [1]

What one API job costs

The standard API rates, in dollars per million tokens, are $8 for input, $0.40 for cached input and $40 for output with gpt-6.2-astra. With gpt-6.2-sol they are $2, $0.10 and $10 respectively. Astra’s ordinary input and output rates are 20% lower than its predecessor’s. Sol retains GPT-6.1 Sol’s rates.

Suppose a job uses 100,000 uncached input tokens and 10,000 billable output tokens. Astra costs $0.80 to read that input and $0.40 to produce the output: $1.20 in total. Sol costs $0.20 plus $0.10, or $0.30. This arithmetic holds token counts fixed and excludes tool charges. If one model finishes in fewer tokens, the ratio between the two bills changes.

Sources: OpenAI Developers [2] · OpenAI Developers [3] · OpenAI [1]

Who gets Astra, Sol and Ultrafast

OpenAI lists Plus, Pro, Enterprise and Edu for the two models in ChatGPT Work and Codex. The faster Ultrafast mode has its own eligibility: Pro 500 subscribers can use it with both models; Pro 200 subscribers get it with Sol. Eligible Enterprise and Edu workspaces need an administrator to enable the option. The API offers Ultrafast for both at launch.

Sources: OpenAI [1] · ChatGPT / X [5]

X / CHATGPT

The rollout from the ChatGPT account

Loading post from X…

Credit / ChatGPT / @ChatGPTOpen original 

ChatGPT singles out making games, slides and websites. The written release lists the plans and modes available.

SOURCES & METHOD

The record behind the article

Checked 17 November 2026. Technical claims are grounded in developer documentation; videos are credited to their creators. Company claims and third-party reviews are attributed in the text; conclusions are editorial analysis. No original comparative testing of these releases was conducted for this article.

  1. Introducing GPT-6.2 Astra and Sol OpenAI · 2026-11-17
  2. GPT-6.2 Astra — model documentation and pricing OpenAI Developers
  3. GPT-6.2 Sol — model documentation and pricing OpenAI Developers
  4. GPT-6.2 release announcement OpenAI / X · 2026-11-17
  5. GPT-6.2 rollout in ChatGPT Work and Codex ChatGPT / X · 2026-11-17