Codex Error: gpt-5.5 Model Is Not Supported? Causes and Solutions

If you recently opened Codex and ran into the "gpt-5.5 Model Is Not Supported" error, don't panic — it's not a problem with your subscription, nor a product failure. The real reason is that OpenAI retired GPT-5.5 from the ChatGPT account login system on October 14, so when Codex signs in through ChatGPT, it naturally can no longer access that model.

What Does This Error Actually Mean?

Codex supports two main authentication methods: ChatGPT account login and API Key login. Previously, the ChatGPT login path allowed users to select GPT-5.5 as the underlying model, and many users had grown accustomed to that combination.

On October 14, OpenAI removed GPT-5.5 from the ChatGPT login system, which directly means Codex no longer offers GPT-5.5 as a selectable model in ChatGPT account mode. When you manually specify that model, or when an old configuration still contains gpt-5.5, Codex will throw the "model is not supported" error.

In short: the model itself hasn't disappeared — it's just that the ChatGPT login path no longer supports it.

To understand why the same model behaves differently across login methods, you need to understand OpenAI's authentication tiering structure. ChatGPT account login goes through the consumer-facing subscription system, where the list of available models is controlled centrally by OpenAI and adjusted as products evolve. API Key login, on the other hand, goes through the developer-facing platform system, where model versions and lifecycles are relatively independent and typically retained for longer. This "dual-track" setup means the same model name (such as gpt-5.5) can have different availability statuses and retirement timelines across the two paths. So when a model is retired from the ChatGPT login system, the API path may still be available — this isn't a contradiction, but rather a result of different lifecycle management strategies for the two product lines.

Solutions: Switch Models or Switch Login Methods

Option 1: Switch to a Model Supported by ChatGPT Login

If you want to continue using Codex with a ChatGPT account login, the most straightforward approach is to change your model configuration to a currently supported one. In ChatGPT login mode, Codex provides a list of available models — simply choose one to restore normal operation.

Keep in mind that different models vary in reasoning ability, speed, and context handling, so you may need to readjust to the output style after switching.

Option 2: Switch to API Key Login

If your workflow heavily depends on GPT-5.5's characteristics, you can switch to API Key authentication. When accessed via API, GPT-5.5 is still available (depending on your API permissions and quota).

The trade-off is that you'll need to manage API billing separately, but for users who need fine-grained control over model versions, this is a more reliable long-term solution.

The Long-Thread Compaction Trap: A Second Pitfall That's Easy to Overlook

Beyond the model error itself, there's a more subtle issue — the long-thread compaction trap.

During extended Codex sessions, the system compacts historical conversations to control context length. If the compaction process still references a retired GPT-5.5 configuration, it can cause abnormal session states, manifesting as inconsistent behavior after switching models, or loss of critical information in the compacted context.

The practical impact: even after you switch models, old long sessions may still carry residual configuration from the GPT-5.5 era. You'll need to start a new session or manually clean up old threads to fully avoid this.

A context window is the maximum number of tokens a model can process in a single call — exceeding it triggers compaction or truncation. Long-thread compaction is the process by which Codex, when a session's tokens approach the limit, summarizes earlier conversations and discards redundant content to free up space. Compaction is typically incremental, and the resulting summary continues to participate in subsequent reasoning as new context. If the summary was generated while still referencing configuration metadata from a retired model (such as model identifiers or capability tags), switching models afterward can cause a mismatch between context and model capabilities, manifesting as inconsistent response styles before and after, or critical constraints being ignored in the summary. Therefore, when encountering such session anomalies, starting a new session is often more reliable than patching an old thread.

What Else Is Unaffected?

The good news is that the impact of this change is limited:

• Codex Core Features Work Normally: Core capabilities like code generation, completion, and refactoring are unaffected — only the underlying model options have changed.
• API Path Is Unaffected: Using GPT-5.5 via API Key is still possible.
• Other ChatGPT Login Models Work Normally: Only GPT-5.5 was removed; the remaining available models are unaffected.

Quick Troubleshooting Checklist

When you encounter this error, follow these steps:

1. Confirm whether your Codex is using ChatGPT login or API Key login.
2. If it's ChatGPT login, check whether your model configuration still says gpt-5.5.
3. Switch the model to one supported by ChatGPT login, or switch to API Key.
4. If the problem persists, try starting a brand-new session to avoid residual effects from old-thread compaction.
5. Check whether any team-level or project-level shared configurations have gpt-5.5 hardcoded.

This change is essentially a product iteration, not a failure. By understanding the authentication tiering logic behind it, you can quickly pinpoint and resolve the issue.

One-line Summary

Codex's gpt-5.5 unsupported error is caused by OpenAI retiring the model — switch models or use an API Key to fix it.

Paragraph Summary

This article explains the real reason behind Codex's "gpt-5.5 Model Is Not Supported" error: OpenAI retired GPT-5.5 from the ChatGPT login system on October 14, rather than a subscription or product issue. It compares the two authentication paths — ChatGPT account login and API Key login — noting that GPT-5.5 remains available only via the API path. Solutions include switching to a ChatGPT-login-supported model or switching to API Key login. The article also reveals the easily overlooked long-thread compaction trap: old sessions may retain GPT-5.5 configuration causing abnormal behavior, so starting a new session or cleaning up old threads is recommended. Finally, it provides a five-step troubleshooting checklist and clarifies that Codex's core features, the API path, and other models are all unaffected.