OpenAI to Shut Down GPT-4o and GPT-4.1 API on Jan 14, 2027

OpenAI’s API deprecations page lists a new entry dated October 14, 2026: on that date, the company notified developers using certain GPT-4 series models that those models will be removed from the API on January 14, 2027. That is roughly three months of notice—shorter than the at-least-six-month notice OpenAI’s deprecations policy lists for generally available models.

The change is an API platform retirement, not something OpenAI documents here as affecting the ChatGPT consumer app. If your production stack still pins gpt-4o-2024-08-06 or other snapshots below, you have a fixed deadline to retarget traffic, re-run evals, and audit fine-tunes.

What’s happening

Under the heading “2026-10-14: GPT-4 series model deprecations,” OpenAI states: “On October 14, 2026, we notified developers using certain GPT-4 series models of their deprecation and removal from the API on January 14, 2027.”

Deprecation in OpenAI’s terminology means the models are already marked deprecated; at the shutdown date they stop being accessible through the API. OpenAI says impacted customers are notified by email and in documentation, and that you can monitor shutdowns programmatically by calling GET /v1/models (or retrieving a single model) and reading each model’s optional shutdown_date field when present.

The same deprecations page also carries other upcoming retirements—for context only—including October 1, 2026 announcements that gpt-5.3-codex, gpt-5.1, and gpt-5.4-nano will leave the API on April 1, 2027, with recommended replacements gpt-6-sol and gpt-6-luna respectively. Those are separate timelines from the January 14 GPT-4 family shutdown.

Which models and snapshots shut down

Four model families are listed, spanning five snapshot IDs:

GPT-4o has been a widely used, capable multimodal workhorse in the API; GPT-4.1 and its mini variant filled similar roles for teams that wanted newer snapshots without jumping to the latest GPT-6 lineup. After January 14, calls that still reference these IDs should fail once OpenAI completes the shutdown.

The replacement: gpt-6-luna

OpenAI recommends one successor for every row above: gpt-6-luna. On its model page, OpenAI describes GPT-6 Luna as its “most efficient model for focused, high-volume tasks,” with a 1,050,000-token context window, text and image input, and broad Responses API tool support (function calling, structured outputs, web search, computer use, MCP, and more).

OpenAI’s Help Center article Migrating from GPT-4.x to GPT-6 Luna calls Luna the recommended starting point for these migrations and advises starting with reasoning.effort set to none, then tuning from representative workloads. It also notes Luna has a lower per-token price than the GPT-4.x models covered by that guide—worth validating on your own traffic because token counts and reasoning settings can change total spend.

Official list pricing on the gpt-6-luna model page (per 1M tokens) is $0.10 input and $0.50 output. For comparison, OpenAI still lists GPT-4o at $2.50 input and $10 output on its GPT-4o documentation—useful context for cost modeling, not a guarantee your migrated prompts will use the same number of tokens.

Fine-tunes and the broader fine-tuning platform

The October 14 deprecations entry adds: “Fine-tuned versions of these models follow the same shutdown schedule as their underlying base models.” Inference on those fine-tunes ends when the base snapshot shuts down.

Separately, OpenAI’s deprecations page documents platform-wide self-serve fine-tuning changes. Among the dates listed there, January 6, 2027 is when “Active existing customers will no longer be able to create new fine-tuning jobs on this date.”—while inference on existing fine-tunes continues until the underlying base model is deprecated. If you still train on GPT-4 family bases, that January 6 cutoff may matter before the January 14 API removal.

Note that the current gpt-6-luna model page lists the Fine-tuning endpoint as not supported for Luna; teams relying on custom fine-tunes need a migration plan that may include prompt changes, retrieval, or a different base model—not a straight swap of the fine-tune ID alone.

What developers should do now

OpenAI’s migration article outlines a practical sequence: baseline on representative requests, evaluate Luna with reasoning effort none, tune prompts and settings, then roll out gradually while monitoring quality, latency, errors, and cost.

For engineering teams, a concrete checklist looks like this:

• Inventory model strings. Search repos, infra configs, feature flags, and observability dashboards for gpt-4o, gpt-4.1, and the exact snapshot IDs above—not just aliases.
• List fine-tunes. Map custom model IDs to their base snapshots and confirm shutdown dates via the Models API or the deprecations table.
• Run regression evals on gpt-6-luna. Prioritize structured outputs, tool and function calls, multimodal inputs if you use them, and latency-sensitive paths.
• Update defaults before the deadline. Staging first, then a partial production cutover, avoids a single January 14 cliff.
• Re-read official docs. Keep the deprecations page and Help Center migration guide as source of truth if dates or replacements change.

Why it matters

Model deprecations are routine on fast-moving API platforms, but GPT-4o and GPT-4.1 still anchor a large share of production integrations—chatbots, document pipelines, agent tool loops, and legacy Assistants or Chat Completions code paths. A single recommended replacement simplifies OpenAI’s messaging but puts the burden on developers to prove Luna matches behavior their users expect, especially where fine-tuning, long context, or strict output schemas were tuned over months on GPT-4 snapshots.

Three months is enough time for disciplined teams to migrate; it is tight for organizations with hundreds of microservices or opaque third-party wrappers that hide the underlying model ID. Treat January 14, 2027 as a hard API cutoff—and January 6, 2027 as a related milestone if you still create new fine-tuning jobs on older bases.

Sources

• OpenAI API documentation: Deprecations (2026-10-14 GPT-4 series entry)
• OpenAI Help Center: Migrating from GPT-4.x to GPT-6 Luna
• OpenAI API documentation: GPT-6 Luna model page (pricing and capabilities)
• OpenAI API documentation: GPT-4o model page (pricing reference)