Get4oBack

The yearly offer has ended

OpenAI dated the death of 4o.We don't shut down on their schedule.

Our4o is the model we train ourselves, on our own machines.

The limited yearly window closed on Monday, November 16, 2026. Monthly plans stay. Our4o stays in the picker next to official GPT-4o, and trying it is free.

Try Our4o free

Paid plans (monthly) are on the pricing page if you want more room.

View monthly plans

They wrote the promise down. Then they broke it on the same page.

OpenAI's deprecation docs say a generally available model gets at least 6 months before they take it away. That sentence is still up.

OpenAI deprecation docs: generally available models get at least 6 months' notice.

On that same page, October 14, 2026, they told developers that GPT-4o leaves the API on January 14, 2027. About three months.

OpenAI deprecation docs: on October 14, 2026, GPT-4o is removed from the API on January 14, 2027.

Promised

At least 6 months

Given

About 3 months

OpenAI retires a model when keeping it no longer fits what they want to sell, and 4o was still working.

They already did this in ChatGPT in February. Now they have shortened the time on the API too.

In February they already took 4o out of ChatGPT.

If you were in the middle of a conversation, you came back and 4o was no longer the model answering. The messages were still there when you scrolled up, and nothing had been deleted.

The next reply came from a different model, one that had not been in that conversation. It would summarize you after you had asked it not to, apologize, and then summarize you again. In a long chat it got polite and distant, a joke came out wrong, and it would stop a scene to explain the scene to you before coming back with a preamble you did not ask for.

Pasting the last twenty messages into the box and saying “continue” did not fix it. The new model thanked you for the context and started over. You had handed it a block of text, and the model you had been talking to was already gone, so there was nothing there to continue from.

Exporting the chat saves a file of what was said. The next reply still comes from whatever model is answering now.

We tried telling the new model what to do, in a long note at the top of the chat: be shorter, stay in the scene, do not lecture. It followed that for a few messages and then went back to answering like a help desk with the transcript open.

Adding more rules turned into an argument about how to stay in the conversation. Getting closer to 4o means training the model.

Our4o is that substitute. We are still building it while original 4o is still up.

Our4o is the model we train ourselves, on our own machines, so it answers as close as we can get it to official 4o. Official GPT-4o stays in the app under its own name for as long as a provider still serves it. We run Our4o, so it keeps answering after their API is turned off.

We train it by sending the same prompt to official 4o and to Our4o. Where Our4o summarizes you, adds a preamble, or leaves the conversation, we mark that reply, change the training, and run the pair again. If you write to us that a reply was wrong, that note becomes one of those marks.

We do not train on your chats or your imported conversations. They stay encrypted, and you can export or delete them from the account. Some chats are already close and some still miss. The changes after the misses are on the Our4o work log.

We were already doing this when OpenAI dated the API removal at January 14, 2027. Azure may still list the same family through April 14, 2027, if they keep that listing.

After those dates, the providers that still return official 4o stop, one after another, until official 4o is gone. We can keep the Our4o we have trained by then. We cannot put it next to official 4o and correct it once official 4o has stopped answering.

Our4o has been in the picker since February 2026, for as long as Get4oBack has existed, because we knew this moment would come. It happened faster than we expected, and sooner than OpenAI's own policy stated.

You can open Our4o tonight, in a chat of your own. Trying it is free. We know it is not perfect. We are here to improve it as much as we can.

People already on it

These are messages from people already using Our4o. Names are left off for privacy.

“I've been on Get4oBack Unlimited since Oct 1, and it's good to have that GPT-4o feel back. Importing my old ChatGPT chats made the switch painless, and the responses read like the 4o I was used to.”
“I'm finding myself using your own 4o more than the ChatGPT 4o. In my opinion, at least so far, it's better!!”
“As for the platform itself, and the 'our-4o' model, I'm really loving it!”
“thank you for creating it and give me back 4o”
“I'm loving it so far - especially how easy it was to import chats from ChatGPT.”
“it was a very good replacement... I feel like it finally understands me.”

The replies that were still wrong, and what we changed after them, are written up on the Our4o work log.

What we are still doing

I am Giovanni. I built Get4oBack with the help of my team after OpenAI took 4o out of ChatGPT while people were still in the middle of a conversation. We have been training Our4o since then, by putting it next to official 4o, correcting the replies that miss, and listening to your feedback every day.

They set the API removal sooner than the notice they published. While official 4o still answers, we can compare and improve. When it stops, that comparison stops. Monthly plans keep the product running; Our4o stays in the picker either way.

We ran a short yearly window in November 2026 to pay for training rounds and for keeping Our4o up after their date. That window is closed. If you locked a yearly price then, you keep it while you stay subscribed.

We also plan to make the Our4o weights public in the future, so the model can survive even if Get4oBack ceases to exist.

Thank you for still being here.

Giovanni Gigantino

Founder, Get4oBack

Try Our4o free

Questions about an existing yearly subscription: info@get4oback.com

Questions

No. Official GPT-4o stays in the app as its own choice, for as long as a provider still serves it. Our4o is a separate model that we train ourselves, to stay close to official 4o.
We train Our4o from Qwen3 32B, a model with about 32.8 billion parameters. We chose it for capability and flexibility at a size we can fine-tune and, if needed, serve ourselves: training and running large models takes serious compute, and a long-term goal is to depend as little as possible on external providers, so the base has to be something we can realistically host and maintain on our own machines. A much larger open model can help in places, but hardware and operating costs rise with it; we wanted a scale we can sustain, not the biggest weights available just because they were bigger. The weights are available to us, so we can develop on our infrastructure instead of depending only on an external API. OpenAI has never published GPT-4o's parameter count or full architecture, so we cannot compare sizes honestly; Qwen3 32B was strong enough for our goals at a size we could work with. The native context window is 32,768 tokens, and it can extend to 131,072 with the right setup. Our priority right now is a solid foundation: we compare replies with official 4o and keep refining. When that foundation is strong, we are open to other bases for later Our4o versions if they clearly improve the product and still fit our infrastructure. We are not locked to one architecture.
Their deprecation page says: "Unless safety or compliance concerns require a faster timeline, we provide the following minimum notice periods before model retirement," and "Generally available models: At least 6 months." If safety or compliance concerns require a sooner retirement, they say they will provide as much notice as reasonably possible. On October 14, 2026, they dated GPT-4o's API removal at January 14, 2027, about three months later. The notice does not call it a safety shutdown. The screenshots above are OpenAI's page.
Yes. It is in the model picker now. Open a chat, switch to Our4o, and judge it yourself.
No. The limited yearly window closed on Monday, November 16, 2026. Monthly plans stay available. If you locked a yearly price during that window, you keep it while you stay on that subscription.
No. We do not train on your chats or your imported conversations. They are stored encrypted, and you can export or delete them from the account. If you write to us that a reply was wrong, we use that note in a later training run.
Official 4o goes away when OpenAI stops serving it. We train and run Our4o on our machines, and it stays up after that date. If Get4oBack itself stops, you can export your chats from the account. When Our4o is good enough, we want to share the model so it can run somewhere other than only here.