If your production code still calls a GPT-4o or GPT-4.1 snapshot, you are on a ticking clock. On October 14, 2026, OpenAI officially notified developers that these models will be removed from the API on January 14, 2027. This leaves a mere 92-day window to migrate, a timeline that feels dangerously short for systems built on stable, legacy model aliases. The designated replacement for every affected model is gpt-6-luna, a shift that introduces new reasoning defaults and breaks existing fine-tuning workflows.

The Deprecation List and Hard Deadlines

The deprecation notice targets four specific model families: GPT-4o, GPT-4o mini, GPT-4.1, and GPT-4.1 mini. Crucially, this applies to dated snapshots such as gpt-4o-2024-08-06 and gpt-4.1-mini-2025-04-14, not just the generic aliases. OpenAI explicitly stated that fine-tuned versions of these models share the same shutdown schedule, meaning any custom classifier or router built on a fine-tuned 4o mini goes dark on the same day. Developers using Azure must note that Microsoft’s retirement schedule differs, listing an April 14, 2027 deadline, though this page has not yet been updated to reflect OpenAI’s latest notice.

GPT-6-Luna Changes the Game

Migrating to gpt-6-luna is not a simple string swap. While Luna offers lower per-token costsβ€”$0.10 input and $0.50 output per million tokens compared to GPT-4o’s $2.50/$10.00β€”it behaves fundamentally differently. Luna reasons by default at a 'medium' effort level, whereas GPT-4o mini responded immediately. This adds latency and changes output consistency for tasks like ticket classification. Furthermore, Luna does not support fine-tuning, leaving teams with custom-trained 4o mini models without a direct equivalent. The only way to mimic the old behavior is to explicitly set reasoning.effort to 'none' in the API call.

Notice Periods and Policy Inconsistencies

OpenAI’s own deprecation policy states that generally available models receive at least six months' notice, while specialized variants get three. The 92-day window for GPT-4o and GPT-4.1 falls into the latter category, but the lack of a migration guide or explanation for the accelerated timeline raises eyebrows. In contrast, newer models like gpt-5.1 received a six-month runway. This inconsistency forces engineering teams to treat model IDs as critical dependencies with unpredictable lifecycles, requiring rigorous version pinning and automated checks rather than relying on vendor stability.

Key Takeaways

  • GPT-4o, GPT-4o mini, GPT-4.1, and GPT-4.1 mini are removed from the OpenAI API on January 14, 2027.
  • The replacement model, gpt-6-luna, does not support fine-tuning, breaking workflows for custom-trained models.
  • Luna reasons by default, requiring developers to explicitly disable reasoning to match the speed of previous mini models.
  • Azure OpenAI users have different retirement dates, currently listed as April 14, 2027, but this is subject to change.

The Bottom Line

Ninety-two days is a brutal deadline for enterprise-grade migrations, especially when the replacement model lacks feature parity for fine-tuning. Treat your LLM dependencies like critical infrastructure libraries, not black-box utilities.