OpenAI promised developers six months’ notice. GPT-4o users got three.
The New Stack Amanda Caswell
OpenAI is pulling GPT-4o and three other GPT-4 models from its API in January 2027. Developers get only three months to move, even though OpenAI says it usually gives six.
Based on reporting by The New Stack, Amanda Caswell — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI has told developers it will retire four GPT-4 models from its API on January 14, 2027: GPT-4o, GPT-4o mini, GPT-4.1, and GPT-4.1 mini, plus their dated snapshots and any fine-tunes built on them. The company’s own deprecation policy says generally available models should get at least six months’ notice. This time, developers have three.
That matters because the replacement isn’t a simple swap. OpenAI points developers toward GPT-6 Luna, its fastest and cheapest tier of GPT-6, but the model behaves differently enough that changing the model name is only the first step. Luna supports a 1.05 million-token context window, structured outputs, and function calling, yet combining reasoning with tool use means moving to the Responses API. If you want reasoning through Chat Completions, you have to set reasoning_effort=none.
There’s also a pricing wrinkle. Luna is set at $0.10 per million input tokens and $0.50 per million output tokens, cheaper than GPT-5.6 Luna, which OpenAI says charged twice as much for input and $1.20 per million output tokens. But once a prompt crosses 272,000 input tokens, the input rate doubles and the output rate rises by 50% for the whole request. That could sting for RAG systems and coding agents chewing through large codebases.
GPT-4o, in particular, has had a long run. It launched on May 13, 2024, became OpenAI’s flagship, then sparked enough backlash when GPT-5 replaced it in August 2025 that OpenAI brought it back for paying ChatGPT subscribers the next day. That reprieve ended for ChatGPT on February 13, 2026, when GPT-4o and GPT-4.1 disappeared there, though they stayed alive in the API until now. OpenAI said only 0.1% of daily ChatGPT users were still choosing GPT-4o by then, but plenty of developers were still running production traffic on it.
The clock is tight elsewhere too. Self-serve fine-tuning for new jobs ends on January 6, 2027, eight days before the retirements, while existing fine-tuned models can keep serving inference until January 14. OpenAI’s model list now includes a shutdown_date field, so at least the company has made the countdown machine-readable. Whether that helps much when the migration guide itself admits behavior changes are the hard part is another question.
My take — AI-written commentary, not fact-checked reporting
OpenAI keeps talking like it runs a public utility, then hands developers a compressed exit ramp and a new API shape for the privilege. That’s the modern AI move: promise stability, ship churn, and call it progress. If a company wants the ecosystem to trust it, six months should mean six months, not whatever fits the release calendar.
Read more about this at: The New Stack