AI BUILDING GUIDE
GPT-5.5 is leaving ChatGPT—not the OpenAI API.
What the October 14 retirement means for builders, why your Codex sign-in matters, and what API access actually costs.
By Nansen Systems ·
Sources checked:
A retirement notice is not the end of every connection
If you use AI to build a business website or application, a model-retirement headline can sound like your project is about to disappear. The important question is not just which model you use—it is how you access it.
This original Nansen guide was inspired by the OrcaRouter article linked below. We checked the core availability and billing details against official OpenAI documentation on 7 October 2026. It is an explanation, not a promise of permanent availability.
1. What changes on October 14?
OpenAI says GPT-5.5 will retire from ChatGPT, ChatGPT Work and ChatGPT-authenticated Codex on 14 October 2026, across all plans. That retirement does not apply to the OpenAI API. The same model name can therefore have different availability depending on the service and sign-in method.
If you use Codex with your ChatGPT account, choose a replacement available to your plan and client before the deadline. Review saved model selections, workspace defaults, agents, schedules and scripts that still request GPT-5.5. Do not assume a saved setting will migrate itself.
2. Why an API key is different
An API key connects usage to an OpenAI Platform organization rather than included ChatGPT plan credits. API-key-authenticated local Codex workflows use the API path. Check your active authentication method before deciding whether the retirement affects your setup.
This is not a way to receive unlimited work for free. OpenAI bills API-key usage separately at API rates. Some features requiring ChatGPT workspace access or cloud services are limited or unavailable with this authentication method. An API key is not a substitute for every subscription feature.
3. Budget for the API—not just the chat subscription
At our check, OpenAI lists standard GPT-5.5 text pricing at US$5 per million input tokens, US$0.50 per million cached input tokens, and US$30 per million output tokens. Prompts exceeding 272,000 input tokens have higher session rates. These are model charges, not an all-inclusive website hosting price.
Our recommendation: measure the cost of a small, useful task before expanding. Repeated context, long outputs, tool calls and retries can change the bill. A model remaining available does not make it the most economical choice for every job.
4. A practical checklist for Nansen learners
First, identify where the model runs: your ChatGPT subscription, your own API account, or a third-party provider. Second, record the model ID and authentication method. Third, test a small representative task and check both the result and recorded usage. Fourth, keep a tested replacement plan rather than waiting for an outage.
Building with Codex and running AI inside your finished application are separate decisions. You can change the coding assistant without automatically changing the application’s runtime model. Keep your project source and original knowledge files backed up, and avoid making a single model name the identity of your entire business system.
Nansen’s perspective
For a business, continuity means knowing which service you depend on and testing a replacement before you need it. We recommend choosing models by the quality and cost of accepted work—not by headlines alone. This article does not change any Nansen worker’s model, billing account or permissions.
Keep people at the center
Students should use their own authorized accounts and understand who pays for each connection. Never paste a private API key into a class chat, shared document or public website. Keep human review for sensitive decisions, and verify the latest provider notices before changing a working setup.
The takeaway
Check the access path before you panic: ChatGPT sign-in and API-key usage are different, and API access carries a separate bill. Available today does not mean guaranteed forever.
Put the ideas to work.
Learn to build your company’s AI Twin with us, or have Nansen build it for you.