OpenAI is retiring GPT-4o from its API on January 14, 2027. On October 14, 2026, it notified developers that GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini are deprecated, per its deprecations page. The named replacement for all four is GPT-6 Luna.
If an app or pipeline still calls one of these models, it has three months to move. Here is what changes, what the replacement costs, and what to check first.
On this page · 6 sectionsOpen
OpenAI notified developers on October 14, 2026 that four GPT-4 series models leave the API on January 14, 2027.
The four: GPT-4o (two snapshots), GPT-4o mini, GPT-4.1 and GPT-4.1 mini; fine-tuned versions follow the same schedule.
OpenAI names one replacement for all four: gpt-6-luna.
On OpenAI’s model pages, GPT-6 Luna lists $0.10 input and $0.50 output per million tokens, against $2.50 and $10 for GPT-4o.
The notice runs three months; OpenAI’s own policy promises generally available models at least six, with an exception for safety or compliance.
§ 01What OpenAI announced
The notice is short: “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.”
| Model | Snapshots | Shutdown | Replacement |
|---|---|---|---|
| GPT-4o | gpt-4o-2024-08-06, gpt-4o-2024-11-20 | January 14, 2027 | gpt-6-luna |
| GPT-4o mini | gpt-4o-mini-2024-07-18 | January 14, 2027 | gpt-6-luna |
| GPT-4.1 | gpt-4.1-2025-04-14 | January 14, 2027 | gpt-6-luna |
| GPT-4.1 mini | gpt-4.1-mini-2025-04-14 | January 14, 2027 | gpt-6-luna |
One line matters most for teams that trained their own versions: “Fine-tuned versions of these models follow the same shutdown schedule as their underlying base models.” A fine-tune on GPT-4o stops on the same day as GPT-4o.
§ 02The notice is shorter than OpenAI’s usual
OpenAI’s own policy, on the same page, sets the notice it aims to give: at least six months for generally available models. This notice runs three months, from October 14 to January 14.
The page gives no reason. The policy does carry an exception: “If safety or compliance concerns require us to retire a model sooner, we will provide as much notice as reasonably possible.” We could not find a statement from OpenAI tying this notice to that clause, so we read it only as a shorter window than the policy’s default.
For teams that cannot move in time, OpenAI says “In some cases, developers may be able to provision dedicated capacity for continued access after a model’s shutdown date,” through its sales team.
§ 03What the replacement costs
On price, the move goes in the developer’s favor. These are the list prices on OpenAI’s model pages for GPT-6 Luna, GPT-4o and GPT-4.1, per million tokens.
| Model | Input | Cached input | Output | Context window | Max output |
|---|---|---|---|---|---|
| GPT-4o | $2.50 | $1.25 | $10.00 | 128,000 | 16,384 |
| GPT-4o mini | $0.15 | $0.075 | $0.60 | 128,000 | 16,384 |
| GPT-4.1 | $2.00 | $0.50 | $8.00 | 1,047,576 | 32,768 |
| GPT-4.1 mini | $0.40 | $0.10 | $1.60 | 1,047,576 | 32,768 |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 | 1,050,000 | 128,000 |
At list price, GPT-6 Luna’s input is 25 times cheaper than GPT-4o’s and its output 20 times cheaper. Against GPT-4o mini, the cheapest of the four, Luna is 1.5 times cheaper on input and 1.2 times on output. Luna’s context window is about eight times GPT-4o’s, and its maximum output is nearly eight times larger. Its knowledge cutoff is May 18, 2026, against October 1, 2023 for GPT-4o.
§ 04What to test before you switch
A cheaper, newer model is not automatically the same model. Before January:
- Run your own prompts. Output style, length and formatting can change between model families. Compare a sample of real requests side by side.
- Check structured outputs. If your code parses JSON or tool calls, confirm the parse rate holds.
- Watch latency. A different model can answer faster or slower; measure it on your traffic.
- Rebuild fine-tunes early. If you depend on a fine-tuned GPT-4o or GPT-4.1, confirm what you can rebuild on the replacement before the cutoff.
- Pin the replacement. Swapping the model name is one line; keeping the old and new paths behind a flag lets you roll back during testing.
§ 05What we are watching
- Whether OpenAI explains the three-month window, or extends it.
- Whether ChatGPT changes which models it offers; this notice covers the API only.
- Whether more GPT-4 series snapshots join the list.
§ 06Sources
- OpenAI API deprecations (read October 14, 2026)
- GPT-6 Luna model page
- GPT-4o model page and GPT-4o mini
- GPT-4.1 model page and GPT-4.1 mini
Q1Do fine-tuned GPT-4o models keep working?
No. OpenAI says fine-tuned versions of these models follow the same shutdown schedule as their base models, so they stop on January 14, 2027 too.
Q2Is GPT-6 Luna a drop-in replacement?
It takes the same inputs (text and image) and returns text, with a much larger context window. Prompts, output style and speed can still differ, so test your own workload before switching.
Q3Why only three months of notice?
OpenAI’s page does not say. Its policy lists at least six months for generally available models and allows sooner retirement for safety or compliance concerns.
Q4Can I keep GPT-4o after the shutdown?
OpenAI says some customers may be able to provision dedicated capacity for continued access after a shutdown date, through its sales team.
Q5Is CellCog affected?
No. CellCog’s AI employees run on Claude Opus 5.5, not on these OpenAI models.