By Jonathan West · Updated Nov 18, 2026

How to Use GPT-6.2 in Work, Codex and the OpenAI API

A step-by-step operational guide to choosing your product surface, managing reasoning effort, and setting up Astra or Sol models.

By Jonathan West · Updated Nov 18, 2026

To use GPT-6.2, open ChatGPT Work, access Codex, or call the Responses API. At Layer3Labs, we build custom agents and enterprise integrations for business teams. In ChatGPT Work, choose Astra or Sol in the model controls.

In Codex, use the model picker or the /model command. In the Responses API, specify gpt-6.2-astra or gpt-6.2-sol. This rollout does not add either model to regular Chat, and access depends on your plan, workspace permissions, and rollout status.

OpenAI released the GPT-6.2 model family on November 17, 2026, introducing GPT-6.2 Astra for frontier professional tasks and GPT-6.2 Sol for everyday knowledge and coding work. The rollout follows specific workspace rules, account tier permissions, and technical constraints.


Start GPT-6.2 in the Right Product

To start using GPT-6.2, select the specific OpenAI product environment that supports the model family: ChatGPT Work, Codex, or the OpenAI Responses API. The standard consumer interface, regular Chat, does not include GPT-6.2 Astra or GPT-6.2 Sol under this deployment.

Account entitlement determines which models appear in your workspace. OpenAI supports GPT-6.2 in ChatGPT Work for Plus, Pro, Business, Enterprise, and Education (Edu) subscription tiers. Workspaces on Enterprise and Edu plans require an administrator to enable model access before individual members can select either model.

Availability also rolls out progressively, which means an eligible subscription alone does not guarantee immediate access.

The three supported surfaces divide work between document production, software development, and automated programmatic execution:

Product SurfaceSupported Account TiersPrimary Control MethodDocumented Reference
ChatGPT WorkPlus, Pro, Business, Enterprise, EduModel dropdown or Power slider (Advanced)OpenAI Learn Documentation
Codex App and CLIPlus, Pro, Business, Enterprise, EduApp model picker or /model commandOpenAI Help Center
OpenAI Responses APIRegistered developer project accountsModel identifiers gpt-6.2-astra and gpt-6.2-solOpenAI Developers Portal

Usage tracking also differs across surfaces. ChatGPT Work and Codex share a unified usage pool governed by your subscription plan. In contrast, the OpenAI Application Programming Interface (API) operates as a developer platform with separate project billing, usage tiers, and rate limits.

ChatGPT Work and Codex share usage under your subscription, whereas the API relies on separate project billing and limits.

  1. Check your account plan: Confirm that your organization or personal profile runs an active Plus, Pro, Business, Enterprise, or Edu subscription for user interfaces, or a funded developer project for API access.
  2. Navigate to ChatGPT Work: On web, desktop, or mobile applications, switch from standard Chat to Work above the conversation composer.
  3. Verify administrative enablement: If you operate on an Enterprise or Edu workspace, verify with your administrator that GPT-6.2 models are toggled on in the workspace console.
  4. Choose between Astra and Sol: Pick GPT-6.2 Sol for iterative knowledge work and routine coding, or GPT-6.2 Astra for long-horizon documents, cross-service debugging, and complex data analysis.

Use GPT-6.2 in ChatGPT Work

To use GPT-6.2 in ChatGPT Work, navigate to the Work workspace within ChatGPT on web, desktop, or mobile. Look above the message composer to ensure the active interface is set to Work rather than standard Chat. If your account shows the Power slider, select Advanced to reveal model selection and reasoning effort settings.

Default model behaviors reflect your account tier. Newly created consumer Work chats default to GPT-6.2 Sol with Light reasoning effort. Enterprise accounts with access enabled default to GPT-6.2 Sol with Medium reasoning effort.

Adjust reasoning effort to match the task. Supply the files the answer must use. State the deliverable and acceptance criteria before asking for a draft.

Operating GPT-6.2 in ChatGPT Work is most reliable when tasks are structured around explicit boundary conditions, supplied reference files, and unambiguous criteria for success. The model works through document synthesis, structured tables, and report drafting by evaluating your source files directly.

  1. Switch to the Work product: In the desktop app, web browser, or mobile client, click Work above the composer to exit standard Chat.
  2. Reveal advanced controls: If the interface displays a Power slider, open Advanced to display the active model menu and reasoning effort controls.
  3. Choose the model version: Select GPT-6.2 Sol for drafting, meeting summaries, and daily task analysis, or GPT-6.2 Astra for comprehensive multi-source presentations and formal reporting.
  4. Adjust reasoning effort: Select an effort level if shown. New consumer Work chats default to Sol Light, while enabled Enterprise chats default to Sol Medium.
  5. Attach source files: Upload reference materials directly into the chat thread to ground the model on source facts rather than open-ended recall.
  6. Provide evaluation instructions: State the exact format, audience level, and required sections before executing the generation.

Choose GPT-6.2 in Codex

To use GPT-6.2 in Codex, open the Codex desktop application or launch the Codex Command Line Interface (CLI). Codex is designed for software tasks with codebase context. In the desktop application, select GPT-6.2 from the model dropdown menu.

In the terminal environment, use the /model command to view and select active models.

Updating the application is a prerequisite for seeing the latest model configurations. On desktop, check Menu, select Check for Updates, and restart the app completely. In terminal environments, run your package manager update command to bring the Codex CLI up to date.

Updating the software prepares the client, though actual model access still depends on whether the model has reached your account in the staged rollout.

Work and Codex share the same underlying plan limits. Pro200 accounts receive Ultrafast speeds on GPT-6.2 Sol. Pro500 accounts receive Ultrafast speeds on both GPT-6.2 Sol and GPT-6.2 Astra.

In Enterprise and Edu environments, the workspace owner must explicitly enable Ultrafast routing before users can activate higher throughput speeds.

  1. Update the Codex client: Run Menu > Check for Updates in the Codex desktop app and restart it, or update the Codex CLI through your command line package manager.
  2. Open your repository: Launch Codex within your local codebase directory so the model can inspect project architecture and configuration files.
  3. Switch the active model: Use the interface model picker in the Codex app, or enter /model in the CLI prompt to select gpt-6.2-sol or gpt-6.2-astra.
  4. Assign a bounded coding objective: Request a targeted review of a pull request, a localized bug remediation, or unit test generation with specified mock dependencies.
  5. Inspect generated code diffs: Review the proposed code modifications against project style guides and execute local test suites before committing changes.

Send a GPT-6.2 Responses Request

To send a GPT-6.2 request programmatically, call the OpenAI Responses API using official model identifier gpt-6.2-sol or gpt-6.2-astra. The Responses API accepts both text and image inputs and returns text outputs. Both models support function calling and developer tools, but neither model accepts a reasoning effort of none.

Passing none is rejected by the API.

For new API integrations, OpenAI recommends setting reasoning effort to medium, or using low instead of legacy minimal settings. In the Responses API, legacy sampling parameters like temperature, top_p, and top_logprobs must be omitted when calling GPT-6.2 models. Placing stable developer messages first in the input array supports prompt caching.

For a comprehensive breakdown of token rates, caching discounts, and long-context surcharges, consult our detailed analysis of GPT-6.2 pricing.

The following Python example sets up an authenticated client using the official openai-python SDK and dispatches a standard request to GPT-6.2 Sol:

import os
from openai import OpenAI

# Ensure OPENAI_API_KEY is configured in your environment variables
client = OpenAI(
 api_key=os.environ.get("OPENAI_API_KEY")
)

response = client.responses.create(
 model="gpt-6.2-sol",
 input="Write a project brief from these example notes: draft due Friday, owner unassigned.",
 reasoning={"effort": "medium"}
)

print(response.output_text)

Install the package with pip install openai. Set OPENAI_API_KEY privately in your environment. Save the example as example.py, then run python example.py.

API usage is billed to your OpenAI developer organization on a per-token basis and does not draw from ChatGPT Work plan allowances.

  1. Create an API secret key: Open the OpenAI API Keys Dashboard and generate a new key for your developer project.
  2. Set your environment variable: Export OPENAI_API_KEY privately in your runtime shell rather than hardcoding credentials inside application files.
  3. Configure your project billing: Verify payment methods and spending limits in the OpenAI Billing Overview.
  4. Call the Responses endpoint: Pass model="gpt-6.2-sol" or model="gpt-6.2-astra" with reasoning={"effort": "medium"}.
  5. Omit legacy parameters: Remove temperature, top_p, and logprob configuration flags from your API payload to avoid validation rejections.

Fix a Missing Model Option

If GPT-6.2 Astra or GPT-6.2 Sol does not appear in your interface, work through an operational troubleshooting checklist before contacting support. First verify that you are not checking regular Chat, which does not include these models under this rollout. You must switch explicitly to ChatGPT Work or launch Codex.

Next check your account plan and workspace permissions. Free and Go tiers do not have access to GPT-6.2 models in ChatGPT Work. On Enterprise and Education workspaces, models remain hidden until an authorized administrator toggles them on inside workspace management policies.

If your account holds an eligible paid subscription and your administrator has approved the models, update your desktop app to the newest build and restart the application completely. Updating your desktop application does not guarantee model access if it has not yet reached your account in the rollout.

  1. Confirm you are in ChatGPT Work: Verify that the top workspace selector displays Work, because regular Chat will not display GPT-6.2 options.
  2. Verify subscription tier: Check account settings to confirm an active Plus, Pro, Business, Enterprise, or Edu subscription.
  3. Check admin console policies: If using an Enterprise or Edu seat, confirm with your IT administrator that GPT-6.2 models are enabled for your permission group.
  4. Run application updates: In desktop ChatGPT or Codex, go to Menu > Check for Updates. Install the update and restart the app.
  5. Check CLI packages: In terminal setups, update your Codex CLI package to ensure local command definitions recognize the latest model endpoints.
  6. Separate API issues from user accounts: Remember that API access requires a provisioned developer account on the OpenAI developer platform, independent of ChatGPT consumer plans.

Choose a First Task and Review the Result

To test your setup effectively, select a self-contained pilot task that matches OpenAI's documented use cases for each model. Running a pilot task validates your connection, reveals how reasoning effort impacts generation time, and lets you evaluate quality against known standards before routing production work.

For GPT-6.2 Sol, OpenAI highlights everyday drafting and engineering tasks, such as conducting a Pull Request (PR) review, revising previously reviewed code files, or transforming messy meeting transcripts into structured project briefs with clear deliverables and owners. For an in-depth review of this model's specifications and task suitability, read our guide on GPT-6.2 Sol capabilities.

For GPT-6.2 Astra, OpenAI recommends complex, multi-stage workflows. Documented examples include cross-service bug diagnosis, fixes, and tests, converting lengthy analytical reports into presentation slide decks using provided corporate templates, and synthesizing research papers that present conflicting empirical findings. Learn more about its architecture in our guide to GPT-6.2 Astra capabilities.

Always review the generated output against provided source files and project criteria. For document synthesis, verify that names, numerical figures, and factual citations match uploaded source documents exactly. For code generation, execute automated test suites locally and require developer approval before deploying changes to shared repositories.

This verification workflow is not designed for Free-tier users who require zero-cost models without subscriptions, nor for teams expecting immediate GPT-6.2 availability inside standard consumer Chat.

Our operational guidance would change if OpenAI expands GPT-6.2 models into standard consumer Chat, adjusts reasoning effort parameters in the Responses API, or alters enterprise workspace enablement policies. Check your account entitlement in ChatGPT Work or Codex, and start learning how to use GPT-6.2 with a single bounded document or code review.

  1. Define input boundaries: Supply complete source files, meeting notes, or code diffs directly in your prompt rather than prompting without context.
  2. Match task to model tier: Assign meeting briefs and revising reviewed code to Sol. Assign report-to-deck drafting and cross-service bug diagnosis to Astra.
  3. Cross-check factual claims: Verify that names, dates, financial figures, and technical assertions directly match uploaded reference materials.
  4. Run unit and integration tests: For Codex outputs, execute your local build and test suites before merging or committing generated modifications.
  5. Require human sign-off: Maintain mandatory operational approval before dispatching model-generated reports to clients or deploying code to production branches.

Frequently Asked Questions

  • GPT-6.2 is rolling out exclusively in ChatGPT Work, Codex, and the Responses API, not standard Chat. If you are in Work, verify that your account has an active Plus, Pro, Business, Enterprise, or Edu plan. Enterprise and Edu accounts require an administrator to enable the model in the workspace console. If eligible, update and restart your desktop app, though access ultimately depends on rollout progression.
  • No. OpenAI's official rollout documentation confirms that GPT-6.2 Astra and GPT-6.2 Sol are not added to regular Chat. To use these models within ChatGPT, you must switch your interface to ChatGPT Work above the message composer.
  • GPT-6.2 access in ChatGPT Work and Codex requires an active Plus, Pro, Business, Enterprise, or Edu subscription, while the Responses API uses separate developer project billing. Neither model is available on Free or Go plans or in regular Chat, and API usage is not provided as a free account entitlement.
  • In the Codex desktop application, open the model picker dropdown to choose your model. In the Codex CLI, enter the /model command to view and select between gpt-6.2-sol and gpt-6.2-astra.
  • No. In the OpenAI Responses API, neither GPT-6.2 Astra nor GPT-6.2 Sol accepts a reasoning effort of none. Attempting to pass none is rejected by the API. OpenAI recommends medium effort for new integrations, and low effort instead of prior minimal settings.

Schedule My Review

Talk with Layer3Labs about integrating GPT-6.2 into your business operations and developer pipelines. We review workspace setup, reasoning effort configurations, and automated model routing for business teams.

Schedule My Review