Summary: This article covers three scenarios—personal subscriptions, enterprise seats, and API projects—and explains how to get access to GPT-6.2 Sol, what to check in order when the model doesn’t show up, and which accounts can use Ultrafast.
Bottom line first: To get access to GPT-6.2 Sol, start by choosing the right entry point for the task. Use ChatGPT Work for multi-step work and deliverables, Codex for software development and terminal tasks, and the API for apps you build yourself, with the model ID gpt-6.2-sol. Once you’re in the right entry point, check the signed-in account, workspace, seat, client version, and corresponding billing in that order. Regular Chat keeps its existing model choices, so it’s normal not to see GPT-6.2 Sol in Chat. There’s one more restriction for enterprise accounts: in Enterprise and Edu workspaces, the model is off by default and must be enabled by an owner.

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026. Sol is positioned for frequent day-to-day work, routine coding, and scalable multi-step agents, while Astra targets harder reasoning and professional deliverables. This article only covers how to use Sol and doesn’t go into a capability comparison between the two.
GPT-6.2 Sol Knowledge Map: Identity → Entry Point → Checks → Access Conditions
The table below organizes the article by “who you are.” Find the row that matches you, then jump to the corresponding numbered section.
| Identity | Entry point | Checks if you can’t find it | Access conditions | Section |
|---|---|---|---|---|
| Individual Plus / Pro | Model picker in ChatGPT Work or Codex | Current view, signed-in account, desktop update, CLI version | Initially available as a Work and Codex desktop preview; what the picker actually shows is the source of truth | 1 |
| Business member | Switch to the Business workspace, then enter Work or Codex according to your seat | Signed-in workspace, seat scope, workspace permissions | Standard and Premium seats each have their own usage rules | 2.1 |
| Enterprise / Edu member | Work and Codex desktop preview in an eligible workspace | Whether the owner has enabled it, Models → Test result, inference residency region | Off by default; only US and EU standard inference are supported | 2.2–2.3 |
| API developer | Call gpt-6.2-sol in an API project |
Project/org access, API billing, model ID, endpoint | Responses and Chat Completions; tool calls go through Responses | 3 |
| Users who want lower latency | Speed option in Codex or Work | Plan tier, model, workspace quota | Only Pro 200, Pro 500, and eligible Enterprise/Edu | 4 |
Reading order: Sections 1, 2, and 3 are three parallel paths—check the one that matches your identity. Section 2 builds on the troubleshooting order in Section 1 and adds the workspace and seat layer. Section 3 covers the API, which uses a separate account and quota system from the first two sections; subscriptions can’t be used directly as API quota. Section 4 assumes you can already see GPT-6.2 Sol in the picker, because Ultrafast is a separate speed tier enabled on top of the model. After reading, you should be able to tell which entry point you should use, where you’re stuck, and whether you can select Ultrafast.
Terms to know before understanding the GPT-6.2 Sol entry points
| Term | Explanation |
|---|---|
| ChatGPT Work | The experience in ChatGPT for multi-step work and deliverables |
| Codex | A tool for software development, codebases, and terminal tasks, available on desktop and as a CLI |
| Model picker | The menu for selecting available models in the current product and signed-in account |
| Workspace seat | The subscription or permission type assigned to a member in a workspace, which affects available features |
| Inference residency | A workspace requirement for the region where model processing takes place |
| API project | The unit for managing development keys, project permissions, usage, and invocation settings |
| Ultrafast | A lower-latency speed option available only to certain models and accounts |

1. How to Use GPT-6.2 Sol with a Personal Subscription
1.1 Choose GPT-6.2 Sol from Either Work or Codex
Personal Plus or Pro users have two entry points: the model selector in ChatGPT Work and the model selector in Codex. The official announcement lists the new model as a preview feature on Work and the Codex desktop app, so whether you can actually select it depends on what you see in your selector. Availability of Work on web and mobile depends on your plan and, again, on what your current selector shows.
The most common mistake happens in regular Chat. Regular Chat keeps the existing model selection; the option shown as GPT-6 in the selector and explicitly choosing GPT-6.2 Sol in Work or Codex are two different entry points. If you only look in regular Chat, you’ll probably think you don’t have access.
There’s also a detail about existing conversations. Some Work conversations that previously used Astra or Sol will show an optional upgrade prompt, and you can decide whether to upgrade. New conversations under the corresponding Default option update automatically according to product rules.
1.2 Troubleshooting Order When GPT-6.2 Sol Doesn’t Appear on a Personal Account
I’d check in the following order, and not move on until the previous step passes:
- Confirm the view: You’re on the Work or Codex page; regular Chat doesn’t count.
- Confirm the account: You’re signed in to the paid account. Users who have both a personal and a company account are the most likely to sign in to the wrong one.
- Update the desktop app: Check for updates in the desktop app menu, install it, and fully restart the app.
- Check the CLI version: If you’re using Codex CLI, the version must be at least 0.166.0.
- Check the selector again: If GPT-6.2 Sol still isn’t there after these steps, it means this preview hasn’t been enabled for your account or device yet.
2. Why Company Accounts Can’t See GPT-6.2 Sol
Company accounts have two extra layers of control that personal accounts don’t: workspaces and seats. A member might have their own subscription, but that doesn’t mean the workspace has made this model available to them.
2.1 The GPT-6.2 Sol Path in Business Workspaces
Business members first need to switch to the Business workspace. Then they pick their entry point based on seat type: seats with Work access choose in Work, while Codex-only seats can only choose in Codex. It’s perfectly normal for members on Codex-only seats not to find GPT-6.2 Sol in Work—switching to Codex will fix it.
When troubleshooting, check three things in order: whether you’re currently signed in to the Business workspace, which products your seat covers, and the workspace-level permission settings. Also, Standard and Premium seats each have their own usage rules, so once you can select the model, keep an eye on the usage cap for your seat.
2.2 GPT-6.2 Sol Is Disabled by Default for Enterprise and Edu
Enterprise and Edu have stricter rules. Eligible workspaces can enable GPT-6.2 Sol in Work and the Codex desktop preview, but the model is disabled by default. Workspace owners can turn it on for the entire workspace or only for selected roles.
Here’s a typical scenario. An engineer’s account has a valid subscription, and coworkers are already using GPT-6.2 Sol, but they can’t see it themselves. The reason is that the owner only enabled access for the “R&D group” role, while their account was assigned to a different role. In this case, the member doesn’t need to do anything; the owner can check the member’s actual permissions in the admin console under Models → Test, confirm it’s a role configuration issue, and then add the role they belong to into the enabled scope.
2.3 Inference Residency Can Make the GPT-6.2 Sol Entry Point Unavailable
Even if the owner has enabled it, regional requirements can still block the entry point. The current preview supports standard inference in the US and EU, so if the workspace requires inference to be performed in the UAE, GPT-6.2 Sol can’t be used through these entry points.
Regional verification comes down to two separate things: the officially published list of countries or regions where the product is supported, and the workspace’s own inference residency settings. The former determines whether an account can use OpenAI’s products; the latter determines whether this model can be scheduled within the workspace.
| Enterprise Identity | Who Enables It | Member-Side Check | Admin-Side Check |
|---|---|---|---|
| Business | Workspace permission settings | Workspace, seat type (Work / Codex-only) | Seat scope, Standard/Premium usage rules |
| Enterprise | Workspace owner | Signed-in workspace, whether it’s a desktop preview | Role enablement scope, Models → Test, inference residency |
| Edu | Workspace owner | Same as above | Same as above |
3. Which Account and Quota Should You Use to Call GPT-6.2 Sol?
3.1 GPT-6.2 Sol’s API Model ID and Endpoints
When building your own app, use the model ID gpt-6.2-sol in your API project. According to the November 17 API changelog, it can be called through both the Responses and Chat Completions endpoints. Function calling and built-in tools (file search, web search, code interpreter, image generation, and so on) go through the Responses path. Neither endpoint accepts the none reasoning tier, so check that parameter first when migrating old code.
| Spec | GPT-6.2 Sol |
|---|---|
| Model ID | gpt-6.2-sol |
| Context window | 1,050,000 tokens |
| Max input / max output | 922,000 / 128,000 tokens |
| Input / output modalities | Text and images / text; no native audio or video |
| Endpoints | Responses, Chat Completions |
| Public rate limits (Build / Launch / Grow) | 5,000 / 10,000 / 15,000 RPM; 1M / 4M / 40M TPM |
| Platforms you can call from | OpenAI API, APIYI apiyi.com (subject to your console’s group settings) |
When you need to generate an image, Sol kicks off a tool call as part of the flow and a separate image tool produces the output — Sol itself only outputs text. The limits in the table above are the published tier values; your actual account is governed by your organization and project configuration.
3.2 Subscriptions, API Keys, and Credits Are Three Separate GPT-6.2 Sol Quotas
The easiest thing to get confused about here is billing attribution. When you log into Codex with a ChatGPT account, you’re consuming the quota and billing tied to that subscription. When you switch to an API Key login, you’re billed at API pricing. Eligible paid Pro and Business users can take part in official campaigns to claim API credits — but whether you qualify and what the deadline is depends on the options shown in your own account.
If you can’t invoke GPT-6.2 Sol in your API project, check these in order: whether the project and organization have access to that model, whether API billing is active, whether the model ID is written as gpt-6.2-sol, and whether requests that need tool calling are going through Responses.
3.3 Calling GPT-6.2 Sol through APIYI
As of 2026-11-20, the public model list on APIYI (apiyi.com) already includes gpt-6.2-sol, with two entry points labeled responses and chat, plus Default, SVIP, and Codex-related groups. The API is OpenAI-compatible — you only need to swap the base_url and the Key. Here’s the simplest Chat Completions example:
from openai import OpenAI
client = OpenAI(
api_key="YOUR_API_KEY",
base_url="https://api.apiyi.com/v1" # APIYI OpenAI-compatible endpoint
)
resp = client.chat.completions.create(
model="gpt-6.2-sol",
messages=[{"role": "user", "content": "Summarize the risk points in this requirements document in three sentences"}]
)
print(resp.choices[0].message.content)
View the Responses endpoint example (including usage reading)
from openai import OpenAI
client = OpenAI(
api_key="YOUR_API_KEY",
base_url="https://api.apiyi.com/v1" # Call the Responses endpoint through APIYI
)
# Note: GPT-6.2 doesn't accept the none reasoning level — remove that setting from older code first
resp = client.responses.create(
model="gpt-6.2-sol",
input="Check whether the following SQL has an injection risk and suggest fixes: SELECT * FROM users WHERE id = " + "'" + "{user_input}" + "'"
)
print(resp.output_text)
# Total output tokens (reasoning tokens already included)
print("output_tokens:", resp.usage.output_tokens)
# Reasoning tokens are usually listed as a sub-item
details = getattr(resp.usage, "output_tokens_details", None)
if details is not None:
print("reasoning_tokens:", details.reasoning_tokens)
🚀 Quick start: If you’d like to test GPT-6.2 Sol on your own workload first, sign up at APIYI (apiyi.com) and claim the free trial credit for new users. The same Key also lets you switch to Claude, Gemini, and other models for side-by-side comparison. The groups, capabilities, and quotas actually available to your project are whatever the console shows.
On pricing, the official lineup splits short and long context based on whether a single request’s input exceeds 272K tokens. Here’s how the two public price lists compare (USD per million tokens):
| Billing category | Official short context | Official long context | APIYI short tier | APIYI long tier |
|---|---|---|---|---|
| Regular input | $2.00 | $4.00 | $2 | $3 |
| Cache read | $0.10 | $0.20 | $0.10 | $0.15 |
| 30-minute cache write | $2.50 | $5.00 | $2.50 | $3.75 |
| Output (including reasoning tokens) | $10.00 | $15.00 | $10 | $11.25 |
The APIYI column above comes from the pricing expressions in apiyi.com’s public pricing interface, and it also switches tiers at 272,000 input tokens. Your actual invoice is what counts. Tool calls are billed separately under each tool’s own rules.

4. Which Accounts Can Select Ultrafast for GPT-6.2 Sol
Ultrafast is a speed option, not a separate model. Seeing GPT-6.2 Sol in the selector only means the model is available — the speed tier has to be confirmed separately.
| Account | Ultrafast for GPT-6.2 Sol | Notes |
|---|---|---|
| Pro 200 | Available, Codex only | Exclusive to Sol |
| Pro 500 | Available | Both Sol and Astra |
| Eligible Enterprise / Edu | Available | Also subject to workspace permissions and workspace quota |
| Plus, Pro 100, Business | Not available | This speed option doesn’t exist |
Whether standard speed works also needs to be checked against your own seat and settings. For enterprise admins, enabling Ultrafast consumes workspace quota, so rolling it out role by role is easier to control usage than turning it on for everyone at once.
4.1 Quick Reference for the Current GPT-6.2 Sol Pages
This last table is something you can check against directly when troubleshooting: first figure out which page you’re on, then what you should be checking.
| Page you’re on | What to look at | What counts as normal |
|---|---|---|
| Regular Chat | No need to look here | Only the original options; showing GPT-6 is also normal |
| ChatGPT Work | Model selector, workspace in the top-right | GPT-6.2 Sol appears; Codex-only seats won’t see it |
| Codex desktop app | Model selector, app version | After updating and restarting, GPT-6.2 Sol and the available speed tiers appear |
| Codex CLI | Version number and login method | Version ≥ 0.166.0; ChatGPT login uses the subscription, API Key login uses API pricing |
| Admin console | Role enablement scope, Models → Test, inference residency | Target member passes the test, residency region is US or EU |
| API project console | Model access, billing, rate limit tier | gpt-6.2-sol is callable, limits match the Build/Launch/Grow tier |
| APIYI console | Model group, balance, endpoint | gpt-6.2-sol is available in the selected group, and both responses/chat return results |
Working down this table from top to bottom will basically pinpoint why “GPT-6.2 Sol won’t work for me.” If the goal is to try things out in the app first, you can just use a key from APIYI apiyi.com with the code in this article to verify — once it runs, you can then decide which set of accounts to formally integrate.
Author: APIYI Technical Team
Technical discussion: Feel free to reach out to us on the APIYI apiyi.com homepage; more integration resources are available in the docs.apiyi.com documentation center
