AI, operations and technologyEnglish edition

OpenAI sets an end date for GPT-4o and GPT-4.1 in the API

Automations that still depend on GPT-4o or GPT-4.1 now have a deadline. OpenAI says it will retire several API models on 14 January 2027 and recommends migrating to GPT-6 Luna. Changing the model name will be easy; proving the process still works as expected will not.

OpenAI’s official graphic about retiring API models.
OpenAI’s official graphic about retiring API models.

01

Which models are going away, and when

On 14 October 2026, OpenAI said it would retire two GPT-4o versions — gpt-4o-2024-08-06 and gpt-4o-2024-11-20 — as well as GPT-4o mini, GPT-4.1 and GPT-4.1 mini. Access ends on 14 January 2027, leaving three months to migrate.

The retirement also affects fine-tuned models built on those versions. OpenAI recommends GPT-6 Luna as a replacement, but that recommendation alone does not guarantee that every integration will behave the same way.

The change affects the API; it does not mean every ChatGPT experience disappears on that date. The important check is in applications, internal assistants and automations that send requests to specific model IDs.

02

The risk is not in the call; it is in the process

Replacing an identifier in code may take minutes. A real automation, however, depends on prompts, output formats, tools, context limits, response times and decisions chained to other systems.

A small variation can change a classification, break JSON, trigger an exception or produce an answer another program no longer recognises. Billing, inventory, customer-service and reporting workflows need an end-to-end test, not an isolated comparison of model answers.

Technical compatibility matters too. GPT-6 Luna supports tools and function calling in the Responses API; in Chat Completions, function calling requires the reasoning effort to be set to none. Older integrations may need more than a model-name change.

03

GPT-6 Luna changes the economics of automation

OpenAI presents GPT-6 Luna as its most efficient model for focused, high-volume tasks. Its published standard price is $0.10 per million input tokens and $0.50 per million output tokens, with additional discounts for cached input.

The price makes it worth revisiting automations that were previously too expensive: document classification, field extraction, structured responses or initial issue review. Still, the token price is only part of the calculation.

A cheap model can be expensive if it needs more retries, creates exceptions or increases supervision. The useful metric is cost per correct, verifiable operation, including the human work needed to resolve errors.

04

How to plan the migration

Start by inventorying applications, scripts and automation platforms that call these models. Include the primary model, fallback routes, scheduled jobs and related credentials.

Then run a parallel evaluation with anonymised real-world cases: normal inputs, incomplete documents, unexpected formats and edge cases. Compare accuracy, format, latency, tool use, consumption and the rate of manual intervention.

The safest transition uses a controlled version, comparable logs and a rollback path. Once the new behaviour is validated, teams can expand traffic gradually and monitor errors, cost and deviations during the first weeks.

05

The deadline is also an architecture opportunity

Regular retirements show the risk of embedding one provider or model in every corner of a process. Separating instructions, business rules, validations and model access reduces the work required for the next migration.

A configuration layer can change models without modifying the whole application. Automated tests help check that an update does not alter critical results. Structured outputs with validation turn an unexpected answer into a visible error instead of letting it pass silently.

The 14 January deadline should not be treated as routine maintenance. It is a chance to turn fragile automations into measurable, replaceable systems prepared for technology that will keep changing.

Sources

Where could this apply to your business?

I help teams turn technology into clear, measurable processes, with sensible controls around the work that matters.

← Back to the blog