OpenAI’s general-purpose reasoning model for everyday work handles multistep tool workflows, code writing and testing, and clear document drafting for assistants and teams.
OpenAI
| Item | Specification |
|---|---|
| Input | Text, images |
| Output | Text |
| Context window | 1,050,000 tokens |
| Max input | 922,000 tokens |
| Max output | 128,000 tokens |
| Vision | ✓ |
| Function calling | ✓ |
OpenAI released GPT-6.2 Sol on November 17, 2026. It is a general-purpose reasoning model for everyday coding, document drafting, and workflows that call tools across several steps.
Compared with GPT-6.1 Sol, this version improves multistep task execution and written communication. OpenAI reports a gain of 5.3 points over GPT-6.1 Sol on its internal GDPval v2 evaluation of long-horizon knowledge work.
| Feature | Benefit |
|---|---|
| Long context | Include several reference materials and prior interactions in one request to preserve project constraints. |
| Adjustable reasoning effort | Set different levels of reasoning for routine code changes and tasks that need repeated verification. |
| Structured Outputs | Constrain successful responses to a JSON Schema for downstream classification or extraction workflows. |
| Function calling | Register business functions and run queries using the function and arguments selected by the model. |
| File search | Retrieve references from a configured file collection to answer questions using internal materials. |
| Streaming | Display text as it arrives in an interactive assistant or another interface where users read incrementally. |
Start with medium for a new integration. If your previous setup uses none or minimal, OpenAI’s migration guide recommends switching to low, and this model rejects none.
Check whether the response completed and whether it contains a refusal before parsing your business JSON. Even with strict Schema enforcement, handle refusals, incomplete responses, and API errors, and validate any business rules that the Schema does not express.
Increase effort when the task needs repeated checking or revisions based on tool results and the usual setting falls short on your acceptance criteria. OpenAI recommends Sol at medium for changes to an existing codebase and document drafts, and xhigh for multistep work with testing and revision along the way.
Query GET /v1/models and check the exact model ID and version in the visible catalog. The catalog request requires no Bearer API key, and you can use a matching entry to look up the model’s API reference.
Send a minimal request with the exact model ID returned by the catalog. If its documentation lists Chat Completions, send a simple messages payload to POST /v1/chat/completions, inspect the reply in choices, and check the remaining parameters your app depends on.
Tiered by input prompt tokens (incl. cache): once over the threshold, the whole request is billed at the higher tier.
| Token Type | Short context ≤272K | Long context >272K |
|---|---|---|
| Input | $1.50$2-25% | $3.00$4-25% |
| Output | $7.50$10-25% | $11.25$15-25% |
| Cached input | $0.075$0.1-25% | $0.150$0.2-25% |
| Cache write | $1.875$2.5-25% | $3.75$5-25% |
| Cache write (12 hours) | $3.75$5-25% | $7.50$10-25% |