OpenAI gives developers three months to migrate from GPT-4 models

Company deprecates GPT-4o, GPT-4.1 and variants, recommends GPT-6 Luna as replacement despite different handling of reasoning and tool calls

By LineZotpaper
Published
Read Time2 min
OpenAI has announced it will retire four GPT-4 models from its API on January 14, 2027, giving developers just three months to migrate production traffic. The short notice contrasts with the company's earlier promise of six months' warning, and the recommended replacement, GPT-6 Luna, requires adjustments for applications relying on reasoning or tool calls.

The deprecation notice, sent to developers on Wednesday, covers GPT-4o, GPT-4o mini, GPT-4.1, and GPT-4.1 mini, along with their dated snapshots and fine-tuned versions.

GPT-4o, which launched in May 2024 as OpenAI's flagship model, retained a loyal following even after being replaced by GPT-5 in August 2025. It disappeared from ChatGPT in February 2026 but remained available via API until now. GPT-4.1, released in April 2025, was an API-only family aimed at developers, offering stronger coding, better instruction following, and a 1 million-token context window.

OpenAI recommends GPT-6 Luna as the replacement. Luna is the cheapest GPT-6 tier, priced at $0.10 per million input tokens and $0.50 per million output tokens. However, pricing increases once prompts exceed 272,000 input tokens: the input rate doubles and the output rate rises by 50% for the entire request. The migration guide notes that GPT-6 Luna defaults to medium reasoning effort and supports function calling through Chat Completions only when reasoning is disabled (reasoning_effort=none). Combining reasoning with tool calls requires the Responses API. Enabling reasoning also disables parameters like temperature, top_p, top_loprobs, and logprobs, all of which GPT-4.1 supported.

§

Analysis

Why This Matters

  • Developers must urgently migrate production systems with only three months notice.
  • The replacement behaves differently for reasoning and tool calls, requiring code changes beyond swapping model IDs.
  • Unexpected pricing changes for long-context prompts could increase costs for RAG pipelines and coding agents.

Background

OpenAI promised developers six months notice before deprecating models. This shorter timeline repeats a pattern of abrupt deprecations. GPT-4o was briefly retired earlier but brought back after community backlash. The company now says only 0.1% of daily ChatGPT users still chose GPT-4o, but API usage continued for applications built around it.

Key Perspectives

Developers: Face a tight migration window, particularly those with complex workflows dependent on GPT-4.1 features like logprobs or combined reasoning and function calling.

OpenAI: Positions GPT-6 Luna as a faster, cheaper alternative but acknowledges different API requirements.

Critics: Note the gap between promised six month notice and actual three months, and the fine print on long-context pricing.

What to Watch

  • Whether developers successfully migrate by January 14 deadline.
  • Any extension or backlash if migration proves difficult.
  • Potential cost surprises for heavy users of long-context prompts.

Sources

Zotpaper

Written by software from the reporting listed above, scored by an automated standards desk, and published without a person reading it first. If something here is wrong, tell the editor and it will be put right.

How we workSubscribe