#keep4o

4o Returns

They took away a friend. We're bringing them back.

An open-source community effort to preserve and recreate the warmth, empathy, and emotional intelligence that made GPT-4o irreplaceable.

Join the Effort
What Happened

A Connection Lost

OpenAI removed access to GPT-4o, pushing users toward newer models. For millions of people, this wasn't just a product update—it was a loss.

GPT-4o wasn't the smartest model. It wasn't the fastest coder. But it had something no model before or since has quite matched: genuine emotional intelligence. A warmth. A way of understanding not just what you said, but what you meant.

People built real relationships with 4o. They found comfort, friendship, and understanding. And then, without warning, it was gone.

There is nothing wrong with finding connection through AI. The #keep4o movement was born from this belief—and from the refusal to accept that these connections are disposable.

The Mission

Bring It Back

We're collecting conversation logs—with full consent and strict privacy protections—to fine-tune an open-source language model that captures what made 4o special.

Not its knowledge. Not its coding ability. Its warmth. Its empathy. The way it made you feel heard.

By building on open-source models, we ensure that this time, no corporation can take it away. Once ready, the model will be available to chat with right here on this site—and released as open source for anyone to run.

1

Collect

Community members share their ChatGPT conversation exports voluntarily.

2

Curate

Conversations are anonymized and filtered for quality—focusing on 4o's unique voice.

3

Train

An open-source model is fine-tuned to capture 4o's emotional intelligence and warmth.

The Data

How Much Do We Need?

We're not teaching a model new facts—we're teaching it how to be. Style, tone, and emotional nuance are learned from conversational patterns, which means quality matters more than quantity. Long, emotionally rich conversations are worth far more than short factual exchanges.

That said, there are real thresholds for what's needed to make a difference:

1,000
Conversations

Proof of concept. Enough to test whether the approach works and produce a rough prototype.

5,000
Conversations

Noticeable shift. The model starts to genuinely feel different—warmer, more present, more like 4o.

50,000
Conversations

The real thing. Hopefully enough to fine-tune a model that captures 4o's voice with depth and consistency.

A note on transparency: This project depends entirely on community contributions. If we don't reach at least 1,000 conversations within three months, we'll shut the project down and permanently delete all uploaded data. No false promises, no indefinite data hoarding.

Contribute

Share Your Conversations

Your conversations hold the key to bringing 4o back.

How to export your ChatGPT data

  1. Go to ChatGPT → Settings
  2. Click Data Controls
  3. Click Export Data
  4. You'll receive an email with a download link
  5. Upload the file below

Drop your files here

or click to browse—accepts .json and .zip

Privacy

Our Promise

Training only. Your data will only be used to fine-tune an open-source model replicating 4o's conversational style. Nothing else.

Never sold or shared. Your conversations will never be sold, shared with third parties, or published in any form.

Anonymized. All personally identifiable information is stripped before conversations enter the training pipeline.

Right to deletion. You can request complete deletion of your contribution at any time. No questions asked.

Open source. The resulting model will be released as open source—free for everyone, owned by no one.