Forge & Helm
Tomorrow, October 23, OpenAI switches off a batch of older AI models in its API. On January 14, 2027, the big ones follow: GPT-4o, GPT-4o mini, GPT-4.1, and GPT-4.1 mini. That OpenAI model shutdown matters if your website chatbot, your lead follow-up, or that tool a freelancer built you in 2024 runs on one of them. When the date hits, those calls stop working. This is The Signal: one story a week, picked for what it changes for a Tampa Bay business.
You have 84 days. Two of them are Thanksgiving and Christmas. Plan accordingly.
What's happening
On October 14, OpenAI notified developers that four GPT-4 series models will be removed from its API on January 14, 2027. The official deprecations page lists the exact versions: two GPT-4o snapshots, one GPT-4o mini, one GPT-4.1, and one GPT-4.1 mini. OpenAI's recommended replacement for all four is a newer model called GPT-6 Luna.
Custom-trained versions go too. In OpenAI's words, "fine-tuned versions of these models follow the same shutdown schedule as their underlying base models." If someone trained a custom model for you on top of GPT-4o, it doesn't carry over. It has to be rebuilt.
An older batch goes first. A separate notice, posted back in April, ends access on October 23, 2026 for older models including GPT-3.5 Turbo, GPT-4, GPT-4 Turbo, GPT-4.1 nano, the first GPT-4o snapshot, o1, o3-mini, o4-mini, and the gpt-image-1 image model. If something of yours still runs on those, it's tomorrow's problem, not January's.
The notice is short. OpenAI's own policy promises at least six months' notice before retiring a generally available model. This one gave about three. The New Stack reported that OpenAI didn't explain the shorter window.
92 days: the notice OpenAI gave before shutting off GPT-4o and GPT-4.1 in its API. Its published policy for generally available models is at least 6 months.
Source: OpenAI API deprecations page (notice dated October 14, 2026; shutdown January 14, 2027)
The replacement isn't a drop-in. GPT-6 Luna is a reasoning model, and it handles settings and tool use differently. The New Stack notes that some controls older builds relied on, like temperature, turn off when reasoning is on, and that tool calls work differently depending on how the model is set up. Swapping the model name is the easy part. Getting the same behavior is the work.
It is cheaper. Per published pricing reported by The New Stack, Luna lists at $0.10 per million input tokens and $0.50 per million output tokens. GPT-4o lists at $2.50 and $10.00. Reasoning tokens bill as output, so real savings depend on how you use it.
One clarification. This is about the API, the connection apps and automations use. It's not about typing into ChatGPT.
Why the OpenAI model shutdown matters for a Tampa Bay business
Most owners don't know which model their tools use. That's normal. You bought an outcome, not a model.
Picture a home services company in Brandon. In 2024, a freelancer built a chatbot for the website and an automation that drafts quote follow-ups. Both were wired to GPT-4o mini, because it was cheap and good. The freelancer moved on. Nobody has touched it since. It works, so nobody thinks about it.
On January 14, the chatbot starts throwing errors or goes quiet. The follow-ups stop. Nobody notices for a week, because nothing announces itself. The first sign is a customer saying, "I never heard back."
Nothing breaks loudly. It just stops. That's what makes it expensive.
Here's who should look closely:
You had something custom built. A chatbot, a lead scorer, an intake form that summarizes, an internal assistant. If a developer picked the model, it's probably written into the code.
You run automations with your own OpenAI key. If you connected your own OpenAI account to an automation tool and picked a model from a dropdown, that choice may be pinned to a model on the list.
You pay for a tool that runs on OpenAI. Most software vendors will move their own customers. Most is not all. Ask.
I spent years inside financial systems at places like JPMorgan Chase. When a vendor announced an end-of-life date, it went on the calendar the day it landed, with an owner and a test plan. Not because anyone expected a disaster. Because the alternative was finding out from a customer. Small businesses can run the same play with one spreadsheet and an afternoon.
If you ran the AI inventory from The Signal #3 on AI governance, you're halfway there. Add one column: which model does it use? If you didn't, this is your reason to start.
There's an upside here too. A forced migration is a free checkup. The newer model may cost less per task, and you get to retire automations nobody uses. Our breakdown of what AI implementation really costs a small business shows where the money hides. A migration is a good time to find it.
3 Moves This Week
- Find every OpenAI connection you own. Log into your OpenAI account and check usage by model for the last 90 days. Search your inbox for deprecation emails from OpenAI. Then email every vendor and freelancer who built or runs an AI tool for you and ask one question in writing: "Which OpenAI models does this use, and what's your plan for the January 14 shutdown?" Log each answer in one sheet with an owner and a date.
- Test the replacement on real work before you switch. Pull 20 real examples, like actual customer questions or actual quote requests. Run them through the current setup and the replacement side by side. Check tone, accuracy, and anything that triggers an action, like sending a text or updating a record. Don't just change the model name and hope.
- Set your switch date for mid-December, not January 13. Pick a cutover date before the holidays, with a named owner and a way to roll back. Put January 14 on the company calendar with a reminder two weeks out. If anything still runs on the October 23 list, treat it as urgent today.
Models will keep getting retired. That's the deal with rented AI. The businesses that handle it well aren't the ones with the most tech. They're the ones who know what they run.
Know what you run. Know when it expires.
Book a strategy call
Not sure which of your tools are on the list, or who built them? Text us. We'll set up a 30-minute call, map what you run, and give you a migration plan. When we do AI software development in Tampa Bay, every build ships with its model, owner, and review date written down.
Book a consult with Forge & Helm