The facts
OpenAI announced GPT-6.2 Astra and GPT-6.2 Sol on 17 November 2026. The company specifically targets rework after the first draft: keeping presentation layouts intact, structuring spreadsheets cleanly and tightening document prose. Astra is positioned for demanding projects, while Sol handles routine work.
For Astra, the announcement highlights presentations, spreadsheets and documents: following a provided template more reliably, respecting professional conventions and writing more directly. OpenAI also points to gains in coding, website design and software use. Sol is designed to complete more tasks within the same usage budget.
Where to find the new models
The rollout covers ChatGPT Work and Codex. OpenAI’s help centre notes that Astra and Sol are not added to the classic Chat model picker. Eligible plans include Plus, Pro, Business, Enterprise and Edu, subject to plan tier, workspace permissions and rollout progress.
API pricing
Both models are also available via the API. Standard pricing per million tokens (up to 272,000 input tokens per request):
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6.2 Astra | $8 | $0.40 | $40 |
| GPT-6.2 Sol | $2 | $0.10 | $10 |
Beyond 272,000 input tokens, input and cached-input rates double and output rates are multiplied by 1.5 across the entire request.
Editorial view
Consider a freelancer preparing a monthly client deck from an existing template and results table. If the model respects both, the time saved is measured in avoided fixes: no misplaced columns, forgotten units or conclusions that clash with the figures.
There is also a documented limitation: in OpenAI’s System Card evaluations where a required file or tool is missing, the models sometimes fail to flag the blocker clearly. Instructions added in Codex reduce these omissions in tests, but are not automatically applied to raw API calls.
Our advice: always ask for the list of sources actually used and any unresolved items alongside the deliverable.