What did OpenAI announce on October 14, 2026?
On October 14, 2026 OpenAI told developers that GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini will be removed from its API on January 14, 2027. The recommended replacement for all four is GPT-6 Luna. After that date, requests to these models stop working.
From notice to shutdown
The notice is on OpenAI's deprecations page, which we read on October 30, 2026: "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." Event date: October 14, 2026. Source page checked: October 30, 2026. Any chatbot, agent or automation step that names one of these models needs a change before that date. If your agents were built by someone else and you are not sure what they run on, our AI agent development team can check and move them.
Key dates. Notice: October 14, 2026. Shutdown: January 14, 2027. That is 92 days. Fine-tuned versions of these models "follow the same shutdown schedule as their underlying base models."
Which OpenAI models and snapshots are affected?
Four models, five snapshots: gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14 and gpt-4.1-mini-2025-04-14. Fine-tuned models built on them shut down on the same date.
| Shutdown date | Model | Snapshots named in the notice | Recommended replacement |
|---|---|---|---|
| January 14, 2027 | GPT-4o | gpt-4o-2024-08-06, gpt-4o-2024-11-20 | gpt-6-luna |
| January 14, 2027 | GPT-4o mini | gpt-4o-mini-2024-07-18 | gpt-6-luna |
| January 14, 2027 | GPT-4.1 | gpt-4.1-2025-04-14 | gpt-6-luna |
| January 14, 2027 | GPT-4.1 mini | gpt-4.1-mini-2025-04-14 | gpt-6-luna |
Two related models are already gone. gpt-4.1-nano and the older gpt-4o-2024-05-13 snapshot were shut down on October 23, 2026 under an April 22 notice. OpenAI's GPT-4o mini model page already shows "Status: Deprecated."
What does GPT-6 Luna cost compared with the retiring models?
Less per token in every case. GPT-6 Luna lists $0.10 per million input tokens and $0.50 per million output tokens, against $0.15 and $0.60 for GPT-4o mini and $2.50 and $10.00 for GPT-4o, on OpenAI's pricing page read October 30, 2026.
Cost per 1,000 agent tasks
| Model | Input per 1M | Output per 1M | Cached input per 1M |
|---|---|---|---|
| gpt-4o | $2.50 | $10.00 | $1.25 |
| gpt-4.1 | $2.00 | $8.00 | $0.50 |
| gpt-4.1-mini | $0.40 | $1.60 | $0.10 |
| gpt-4o-mini | $0.15 | $0.60 | $0.075 |
| gpt-6-luna (replacement) | $0.10 | $0.50 | $0.01 |
On the sample workload from our LLM price and retirement tracker (20 million input and 1.2 million output tokens per 1,000 agent tasks), GPT-4o costs $62.00, GPT-4.1 $49.60, GPT-4.1 mini $9.92, GPT-4o mini $3.72 and GPT-6 Luna $2.60. That assumes the same token counts on both models, which may not hold, for the reason in the next section.
What changes when you move an agent to GPT-6 Luna?
GPT-6 Luna is a reasoning model, so it thinks before answering and bills that thinking as output tokens. Its default reasoning effort is medium, temperature is not accepted unless reasoning is off, and in Chat Completions it calls functions only with reasoning set to none.
- Reasoning effort. The model page lists reasoning.effort values of none, low, medium (the default), high, xhigh and max. At none it behaves most like the GPT-4 series.
- Reasoning tokens cost money. OpenAI says reasoning tokens "are billed as output tokens." A per-token price cut can shrink if the model reasons at length, so measure real usage on your own cases.
- Temperature. OpenAI's GPT-6 guidance says "when reasoning effort is not none, remove temperature, top_p, and top_logprobs." Old requests that set temperature need changing. Our page on whether AI agents are deterministic explains what to use instead.
- Tool calling. "Chat Completions supports function calling only with reasoning_effort set to none." For reasoning with tools, OpenAI points to the Responses API.
- Bigger context, newer knowledge. A 1,050,000-token context window and a May 18, 2026 knowledge cutoff, against 128,000 tokens and October 1, 2023 for GPT-4o mini.
None of this means the agent will behave the same. A different model reads your prompt differently, so expect changes in tone, length and tool choices. That is a test question, and the only safe answer is to run your real cases on both models before switching.
Is 92 days the normal notice period from OpenAI?
It is shorter than OpenAI's stated policy. Its deprecations page lists at least 6 months' notice for generally available models unless safety or compliance concerns require faster, yet this notice gives 92 days. OpenAI's other 2026 notices for general models gave 182 to 184 days.
The page states the policy this way: "Generally available models: At least 6 months," "Specialized variants of generally available models: At least 3 months," and preview models "may be retired with much shorter notice, such as 2 weeks." The October 14 notice does not explain why the GPT-4 series gets 92 days. Whatever the reason, the date is the date: plan for January 14, 2027. Our tracker compares notice periods across OpenAI, Anthropic and Google.
What this means for your business
If any chatbot, AI agent or automation step you run names GPT-4o, GPT-4o mini, GPT-4.1 or GPT-4.1 mini, it will stop working on January 14, 2027. Find those steps now, test GPT-6 Luna or another current model on real cases, and switch in December.
- Find every use. Check the model field in each AI step in n8n, Make, Zapier and GoHighLevel, and the model string in custom code. OpenAI's List models endpoint reports a shutdown_date for each model, which a script can check.
- Check fine-tunes. A fine-tuned model on one of these bases shuts down with it.
- Test before you switch. Run your saved test cases on GPT-6 Luna at reasoning effort none and at low, and compare answers, tool calls and token use. Our guide to testing an AI agent before launch covers the method.
- Fix the request. Remove temperature and top_p where reasoning is on, and move tool-calling agents to the Responses API if they need reasoning.
- Switch early. Move by mid-December so there is time to catch problems before the shutdown, and watch the first week of live runs.
Agents built inside GoHighLevel can already use the new models: HighLevel added GPT-6 Sol and Luna to its AI actions in September, as we reported in GPT-6 Sol and Luna arrive in GoHighLevel. New agents we build start on current models, chosen for each client's quality, cost and data needs, so they are not on this list. If we built your agent and it is within 6 months of launch, the fix is free; after that, optional monthly support from $200 covers changes like this. If someone else built it, the free strategy call is a good place to list what needs to move.
More questions about the GPT-4o and GPT-4.1 retirement
Will ChatGPT lose GPT-4o on January 14, 2027?
This notice is about the OpenAI API, the service developers and automation tools call. It does not say anything about the ChatGPT app. If you only use ChatGPT through its website or app, the notice does not tell you what changes there. If a tool or workflow calls OpenAI with your API key, it applies to you.
What happens to requests after the shutdown date?
OpenAI says that at shutdown "the model or endpoint will no longer be accessible." Any request that still names one of these models fails, so an agent, chatbot or automation step that relies on it stops producing answers. Depending on how the step handles errors, that can mean a visible failure or a workflow that silently skips its AI step.
Do we have to use GPT-6 Luna?
No. GPT-6 Luna is OpenAI's recommended replacement, but you can choose any current model, including GPT-6.1 Terra or GPT-6 Sol from OpenAI, or a Claude or Gemini model. Pick by testing your own cases on two or three candidates and comparing quality, speed and real cost per task. Price per token alone can mislead.
Is GPT-6 Luna better than GPT-4o?
OpenAI describes GPT-6 Luna as its "most efficient model for focused, high-volume tasks." It is newer, cheaper per token and has a larger context window and later knowledge cutoff. Whether it does your specific job better is something only a test on your own cases can tell, because a model built for efficiency can trail an older model on some tasks.
Are other OpenAI models retiring soon?
Yes. GPT-5, GPT-5 mini, GPT-5 nano, GPT-5 pro, o3 and o3-pro shut down on December 11, 2026, and GPT-5.1, GPT-5.4 nano and GPT-5.3 Codex on April 1, 2027. OpenAI's Agent Builder, Evals platform and reusable prompts close on November 30, 2026. Our tracker lists every announced date with its replacement.
Sources and further reading
- OpenAI, Deprecations (October 14, 2026 notice, notice-period policy, other 2026 notices), checked October 30, 2026
- OpenAI API pricing, checked October 30, 2026
- OpenAI, GPT-6 Luna model page, checked October 30, 2026
- OpenAI, GPT-4o mini model page, checked October 30, 2026
- OpenAI, Using GPT-6 models (migration checklist), checked October 30, 2026
- OpenAI, Reasoning models, checked October 30, 2026
- OpenAI API changelog (GPT-6 Sol and Luna release, September 22, 2026), checked October 30, 2026
