GPT-6.2 Context Window: API, Work, and Codex Limits
GPT-6.2 Astra and Sol each list a 1,050,000-token context window in the API. The same pages list a maximum input of 922,000 tokens and a maximum output of 128,000 tokens. Codex has its own defaults, and Work has its own file and workspace rules. Use the limits of the entry point where you will actually do the work. Astra API specifications · Sol API specifications
What each number describes
| Setting | GPT-6.2 Astra | GPT-6.2 Sol | |---|---:|---:| | API context window | 1,050,000 | 1,050,000 | | Maximum API input | 922,000 | 922,000 | | Maximum API output | 128,000 | 128,000 | | Default Codex context | 272,000 | 500,000 |
The context window describes the model’s working token budget. Input and output caps specify separate bounds. Your request may include instructions, conversation history, supporting files, and material returned by tools. Treat all material the model actually receives as part of the planning budget. API model comparison
Why the Codex defaults are different
OpenAI lists 272,000 tokens for Astra and 500,000 for Sol as default Codex contexts. Its model guide says both support a Codex context of up to 1 million tokens, with longer settings configured through model_context_window. A client default answers what a normal session starts with, while the API model card describes API limits. Confirm active settings in the environment you use before estimating what will fit. Codex context settings
For Work, use the model options available in your workspace and check whether the files you expect are actually accessible in the session. The official model page presents those Codex figures as Codex defaults. It does not establish a universal Work upload or context limit. A successful upload also leaves a separate question: whether the model used the right part of the file.
Plan a long project around what needs to stay in view
For a video series, begin with a small, reviewed set: the approved story outline, the character reference, and the scenes currently being revised. Ask for a scene-by-scene change list with page or section references. Add earlier material when the work depends on it. At each handoff, retain the approved decisions and record where they came from.
API teams should also plan for the billing threshold. The model pages list higher rates above 272K input tokens for the full request. This is a pricing threshold, separate from the maximum input. Start with a smaller evidence packet and test whether a larger one improves the result enough to justify its cost. Astra API rates · Sol API rates
Frequently asked questions
Can I send 1,050,000 input tokens in one API request? The documented maximum input for these models is 922,000 tokens. Plan below the listed cap and allow room for additional material your application includes.
Does a larger context mean the model will cite everything accurately? Check coverage and citations on a representative long task. Ask for explicit references and verify them against the source documents.




