How to Use GPT-6.2
Find GPT-6.2 by plan, enable Astra and Sol for your workspace, choose a model for your job, and make your first API call.
Published

Open ChatGPT Work or Codex desktop and choose GPT-6.2 Sol for a first pass at everyday work. Choose GPT-6.2 Astra when several sources have to agree in one deliverable, such as a valuation workbook or a slide for an account review. If neither appears in regular Chat, switch to Work first. Buying a different plan is a poor first troubleshooting step.
Where to find GPT-6.2 on your plan
On November 17, 2026, OpenAI released Astra and Sol in ChatGPT Work and Codex. Open the model picker in Work or the model selector in Codex desktop and select the model by name. Regular Chat has its own model choices. The ChatGPT release notes describe the new models as a preview.
Ultrafast is a speed setting. Choose your model first. The Work and Codex access guide explains which accounts can then choose Ultrafast. For a company account, the Business seat guide covers model eligibility. If you're choosing a personal subscription specifically for this feature, Plus is the lowest personal plan the launch names. See ChatGPT Go vs Plus for the rest of that decision.

If Astra or Sol is missing
- Check where you are. Open Work or Codex desktop. Confirm the account and workspace shown are the ones with your eligible plan. If you use Codex from a terminal, the command-line app needs version
0.166.0or later. - Check the member in the admin console. For Enterprise or Edu, select the workspace, open Models → Test, and search by the member's name or email. The result shows which models that person can use and which seat, role or workspace setting applies.
- Enable the model for the people who need it. A workspace owner can turn on the desired GPT-6.2 model in the model access controls for the whole workspace or the relevant role. Save the setting, then use Test to verify that member's access. Test itself only reads the saved settings.
Some workspace configurations have another eligibility requirement. Workspaces requiring processing in the UAE cannot use these previews in Work or Codex desktop. In FedRAMP, Work requires private-preview enrollment as well as model enablement. The Enterprise and Edu model guide covers both cases.
Which model to use for the job in front of you
Start with the cost of a mistake in the finished piece. A shortlist is easy to inspect item by item. A workbook can look polished even when a formula several tabs away points to the wrong cell. Here are three jobs from OpenAI's launch examples, ordered by where I'd spend the extra effort first.

Two more jobs follow the same rule. A product analyst could give Astra the results of a feature experiment and ask for a rollout memo. In OpenAI's fictional Shared Bills example, a lift in use of the feature coexisted with inconclusive results for overall activity. Ask the model to identify which finding actually bears on the rollout decision. A developer handling routine maintenance could give Sol a small proposed change and ask for a test that distinguishes the old behavior from the new one. Run the test before merging. For more on when the extra reasoning earns its keep, see GPT-6.2 Astra vs Sol.
How to call GPT-6.2 from the API
Set model to gpt-6.2-sol or gpt-6.2-astra. Both accept ordinary requests through the Responses API (/v1/responses) and Chat Completions (/v1/chat/completions). If the model needs to call a tool, such as a function that looks up an account record, use Responses. The November API changelog also lists the two-hour and eight-hour background completion windows in the Responses public beta for work that can wait.
With the Python SDK installed and API credentials configured as in the developer quickstart, this request turns a short note into a handoff.
from openai import OpenAI
client = OpenAI()
response = client.responses.create(
model="gpt-6.2-sol",
input="Extract the owner and next step: Jo will send the revised quote Friday.",
)
print(response.output_text)For an existing app, check two settings before swapping the model ID. The effort setting is reasoning.effort in Responses and reasoning_effort in Chat Completions. In Responses, keep reusable guidance in a developer message at the start of the input. Editing the early conversation can discard earlier reasoning and make the model repeat work. Reserve top-level instructions for the current request. The GPT-6.2 migration guide explains the change.
Two tools a team could build
I'd build the weekly account-slide updater first. A sales operations lead could start with one approved slide, one account and one set of dated notes. Have Astra draft the update and flag commitments missing an owner. The lead can review changed fields instead of copying notes into a deck and reformatting it each week. Google's related searches for “sales account plan template” name three formats, Word, PDF and Excel. Before building, ask which file the team actually takes into its review. An attractive slide in the wrong format creates another handoff.
An experiment memo assistant could serve product teams. Give Astra the test plan and results table. Return a one-page draft with the decision metric, uncertainty and proposed next step, ready for analyst review. Test the product with an analyst who already writes these memos. A fluent paragraph can still draw the wrong conclusion from a small sample.
Questions about using GPT-6.2
Why is the model picker missing on my iPhone or Android phone?
The November release identifies the preview in Work and Codex desktop. It does not give a general mobile rollout date. Check an eligible account in Work or Codex desktop before diagnosing a phone problem.
How do I decide which AI model to use?
Pick Sol when you can review the result in one pass. Pick Astra when a decision depends on reconciling sources or matching an existing professional template. OpenAI does not state one message cap that applies to every plan and job.
How should I write up an experiment with GPT-6.2?
Give Astra the original question, the success measure chosen before the test and the observed results. Ask for the answer to that question first and the supporting figures second. Have the analyst verify the interpretation before the memo goes to leadership.
If your team wants that weekly handoff built into its existing tools, see workflow automation services.
- Published
- Category
- AI







