ChatGPT Model Not Found Error Explanations and Fixes

A ChatGPT model not found error occurs when you call an invalid model name, query a retired snapshot, or request a model your account cannot access. At Layer3Labs, we build AI workflow systems for businesses, and diagnosing model routing errors is standard work during our client deployments. The message appears in two distinct environments. You see it inside the web application or as an API response. Both point to specific configuration mismatches.

OpenAI does not formally document a 404 model not found status in its standard API error codes guide. The official guide covers status codes like 400, 401, 403, and 503 instead. That omission leaves developers guessing. They cannot tell whether the issue sits in their request payload, their plan tier, or OpenAI servers.

GPT-5.5 Retirement From ChatGPT on October 14, 2026

ChatGPT returns a model-not-found error when a chat or saved agent still targets GPT-5.5, and Codex returns the same error when its configuration still names gpt-5.5. OpenAI retired GPT-5.5 from ChatGPT, ChatGPT Work, and Codex on October 14, 2026, across consumer, Business, Enterprise, and Edu plans. OpenAI announced the cutoff in a September 14, 2026 changelog entry cited by Mixed and documented the removal in the OpenAI model release notes.

In the web interface, start a new chat and pick a model the selector shows, or leave it on Auto. OpenAI has not published what happens to existing GPT-5.5 conversations or whether custom GPTs pinned to GPT-5.5 switch automatically. If the interface still fails after opening a new session, follow the ordered checks below. Additional details appear in the GPT-5.5 retirement overview and the Codex model not available guide.

Codex and API access depend on how the session signs in:

• Codex accounts signed in through ChatGPT must switch from gpt-5.5 to gpt-5.6-sol across saved model settings, custom agents, scheduled tasks, and scripts.
• Codex sessions authenticated with an API key retain GPT-5.5 access for now, according to an October 15, 2026 press report from 36Kr.
• The retirement does not apply to the OpenAI API, and no GPT-5.5 entry appears on the OpenAI deprecations page as of October 15, 2026.
• A model's shutdown date can be read from the shutdown_date field via GET /v1/models.

Why the API Error Codes Guide Omits Model Not Found

OpenAI omits the model not found error from its standard API error documentation. The official API error codes guide documents other failure codes. It defines code 400 for disallowed service tiers and code 401 for invalid credentials. It outlines code 403 for unsupported regions and code 503 for overloaded models. A missing model triggers a 404 that the guide does not document at all.

The client receives a message stating that the model does not exist or you lack access. That single response conflates two separate operational issues. The string could be misspelled. Alternatively, the model exists within the ecosystem, but your specific organization or project has not received access to it. Checking the raw API payload clarifies which branch you are hitting.

Plan Tiers and Model Rollouts in the ChatGPT App

Access to new models in the ChatGPT application depends directly on your subscription tier. OpenAI announced GPT-6 Astra on September 3, 2026. GPT-6 Pro runs on GPT-6 Astra across the Pro, Business, and Enterprise plans. Plus plans include GPT-6 Astra in ChatGPT Work and Codex during the rollout. Plus users will not see GPT-6 Astra in the primary chat model list.

OpenAI deploys major updates gradually. Availability can differ between Chat, Work, and Codex on the same account. A model your plan cannot select is different from a model that does not exist. If your interface drop-down is missing, read our guide on the ChatGPT model picker missing. That is an interface state. Consult the ChatGPT help article to confirm which models your current tier includes.

Model Deprecations and Scheduled API Shutdowns

Querying a shut-down model returns an error because the model has been retired from service. OpenAI separates deprecation from shutdown. A deprecation is the formal announcement of planned retirement. A deprecated model continues to accept incoming requests and return completions until the shutdown date arrives. It stops answering completely once shutdown occurs.

Retirement notice periods follow fixed schedules. OpenAI provides at least 6 months of notice for generally available models. Specialized variants receive at least 3 months of notice. Preview models can shut down with as little as 2 weeks of notice. Check the OpenAI deprecations page for active timelines.

• gpt-4-0613, gpt-4-turbo-2024-04-09, gpt-3.5-turbo-0125, and o1-2024-12-17 were announced for retirement on April 22, 2026 and are scheduled to shut down on October 23, 2026.
• gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14, and gpt-4.1-mini-2025-04-14 were announced for retirement on October 14, 2026, with a shutdown date of January 14, 2027 and gpt-6-luna as the replacement.
• gpt-5-2025-08-07, gpt-5-mini-2025-08-07, gpt-5-nano-2025-08-07, and o3-2025-04-16 were announced on June 11, 2026 with a shutdown date of December 11, 2026.
• OpenAI designated gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna as replacements for retiring models.

Exact Model Identifiers and SDK Host Routing

OpenAI requires exact string matches for every model identifier passed to its endpoints. The API treats gpt-4 and gpt-4-0613 as two completely distinct identifiers. A single typo in a snapshot date makes the gateway reject the request. Sending a model request to the wrong endpoint also causes immediate failures. A chat model fails on a legacy completions route. Audio models fail on text endpoints.

Host configuration presents another common hurdle. Developers using the official SDK sometimes route requests to Azure OpenAI endpoints. Azure OpenAI resolves custom deployment names instead of base OpenAI model IDs. Passing a valid OpenAI model string to an Azure host returns a resource not found message. Verify your base URL settings if you connect through proxy architectures.

The failure mode we hit most often is teams assuming their API key grants access to every model OpenAI publishes. Keys belong to specific projects. A project often lacks permission for newest preview models. Developers then waste effort rewriting code when the fix requires updating project settings on the OpenAI platform. Check permissions before touching code.

Model Not Found in the ChatGPT App on a Paid Plan

In our reading, the message usually means the current chat or the selected model is no longer available to that conversation. It does not mean your account is broken. An active subscription on its own does not change which model a conversation is tied to.

We suggest following these ordered checks to isolate the issue:

If your model selector is missing entirely, consult the guide on missing ChatGPT model selectors, or see Codex model not available. The chat assistant on this site cannot see or change your ChatGPT account.

• Step 1: Start a brand-new chat instead of continuing the old one, because an old conversation can remain tied to a model that OpenAI no longer offers.
• Step 2: Open the model selector in the new chat and pick from the list shown, or leave the setting on Auto (see which models your tier includes).
• Step 3: Sign out of your account and sign back in, or try a different browser or the mobile app to rule out a stale session.
• Step 4: Check the OpenAI status page to confirm the service is not experiencing an outage.
• Step 5: If your paid plan is active and every new chat still fails, contact OpenAI support through the OpenAI help center with your plan name, device, browser, and the exact error message.

Step-by-Step Triage for Missing Models

A systematic check isolates whether the failure stems from your string, your credentials, or retirement schedules. Run through these checks in order. Each step rules out a specific operational failure.

Review your request logs for specific HTTP response status codes. The status code reveals how OpenAI processed your authentication and project tier. Checking the response headers helps confirm your project routing.

• Step 1: Check the exact model string. Compare your identifier against the active list on the OpenAI platform to catch typos.
• Step 2: Inspect endpoint compatibility. Chat models must target /v1/chat/completions.
• Step 3: Review project permissions. Confirm your API key belongs to an organization and project authorized for that service tier.
• Step 4: Check retirement status. Visit the OpenAI deprecations page to confirm the shutdown date.
• Step 5: Verify host routing. Azure endpoints require deployment names.

Who This Troubleshooting Guide Is Not For

This troubleshooting guide does not address users experiencing broad service outages or missing interface buttons. If OpenAI servers return 500 server errors, the problem sits on OpenAI infrastructure. Code 503 indicates that the requested model is temporarily overloaded. Wait for server capacity to recover when encountering 503 responses.

Users who cannot locate the model selector dropdown in ChatGPT should read our ChatGPT model picker missing guide. That issue involves interface controls. It is not a missing model code error. Developers encountering missing CLI models should consult our Codex model not available article instead.

What Would Change These Troubleshooting Checks

These troubleshooting steps would change if OpenAI adds explicit 404 error definitions to its developer documentation. The missing status code currently creates ambiguity between syntax errors and access limitations. Clear documentation would simplify payload inspection. Our checks would also change if OpenAI standardizes deployment naming conventions across partner platforms.

Which exact models a plan or an API project can reach changes without notice. Verify your project limits directly on the OpenAI platform before refactoring active workflows. Audit your project access permissions as your next troubleshooting step.

How to use ChatGPT

You do not host ChatGPT yourself — you use it through a tool, so "getting started" really means choosing the right one.

The fastest way to put ChatGPT to work day to day is inside an AI IDE, and Cursor is the most popular — it supports it directly, so you can be working in minutes. The maker's own option is Codex for ChatGPT, if you want the native experience. Prefer a different editor? Windsurf, Zed, and GitHub Copilot drive these models too.