GPT-4o Is Retiring From the API: What to Do Before January 14, 2027

OpenAI has announced that the GPT-4o API model will be retired on January 14, 2027. Developers who still use GPT-4o, GPT-4o mini, GPT-4.1, or GPT-4.1 mini after this date will experience failures in their code. This shutdown affects the API only, and users of the ChatGPT interface are not impacted since GPT-4o was removed from ChatGPT's model picker in February 2026.

OpenAI provides a single replacement model, gpt-6-luna, for all four affected models. Custom fine-tunes follow the same schedule as their base models and will also stop on January 14, 2027. The original May 2024 snapshot of GPT-4o will shut down on October 23, 2026, leaving only the August and November 2024 snapshots until January 14, 2027.

The October 14, 2026 notification gives developers three months to prepare for the shutdown. OpenAI has not confirmed whether this notice falls under an exception for models requiring faster deprecation due to safety or compliance concerns. GPT-4o's pricing is significantly higher than the replacement model, with GPT-4o costing $2.50 per million input tokens and $10 per million output tokens, while Luna costs $0.10 per million input tokens and $0.50 per million output tokens.

However, token consumption between models is not the same, so users should test the total cost on their specific workloads. OpenAI's newer models also exhibit reduced sycophantic replies, which may be beneficial for production systems. To prepare for the migration, developers should audit all instances of the affected models in their code, select the appropriate replacement model, test on real workloads, rebuild fine-tunes, validate structured outputs, gradually roll over traffic, monitor costs, and communicate changes to users.