Ajax AI PewDiePie: The Local Model OpenAI Banned Him Over

The Ajax AI PewDiePie launch landed in a new video on October 2, 2026. Felix Kjellberg, better known as PewDiePie, unveiled Ajax, a small AI model built to run on home PCs inside his free Odysseus workspace. He also said OpenAI banned him twice while he was making it. Here is what Ajax is, what it can do and what is still missing.

Ajax is a fine-tuned assistant with 9 billion parameters. According to TBreak, it is based on Alibaba’s Qwen3.5-9B and tuned specifically for Odysseus.

In other words, it is not a giant frontier model. Instead, it is small enough to run on a strong gaming PC, so your chats and files stay on hardware you control.

Kjellberg said the early version completed its intended tasks roughly nine times out of ten. He also called it a work in progress.

Ajax only makes sense next to Odysseus. Kjellberg launched Odysseus in June 2026 as a free, self-hosted AI workspace, pitched as an alternative to ChatGPT and Claude.

The software is open source. Its repository lists the AGPL-3.0-or-later license, and the project is local-first, so conversations and data live on your own machine.

Ajax is the brain built for that workspace. Rather than chatting for its own sake, it is meant to use the tools Odysseus gives it.

In his launch video, Kjellberg says Ajax can search and browse the web. It can also handle email, manage a calendar and work with to-do notes.

That makes it closer to a small always-on agent than a chatbot. For example, you could ask it to find information or tidy up a task list, and it works through Odysseus to do it.

However, these are his claims from the video. Nobody outside his team has tested the model yet, because the weights are not public.

The part of the video that spread fastest was the ban story. Kjellberg says OpenAI suspended his account while he was building Ajax.

An email he showed cited distillation as the reason for the first suspension. In AI, distillation means using one model’s outputs to help train another. According to Free Press Journal, he wanted to use an OpenAI model’s output to seed training data.

He says OpenAI restored the account after an appeal, then banned it again. OpenAI’s terms forbid using its outputs to build competing models. So far, these are his own claims, and OpenAI has not commented publicly.

Here is what is confirmed so far, and what is still unknown.

Meanwhile, the download link pointed to a data contribution page, not to model files.

Kjellberg made Ajax deliberately less restrictive. TBreak reports that he used an open-source tool called Heretic to strip out much of the model’s refusal behaviour.

His goal, he says, is a freer experience. Still, he says he kept limits on requests that could cause harm.

This is the most debated part of the launch. Fans like a model that does not lecture them. Critics, on the other hand, point out that fewer guardrails also mean more room for misuse.

Ajax did not come from nowhere. In 2025, Kjellberg showed off a home AI lab with 10 GPUs, including eight modified RTX 4090 cards with 48 GB of memory each, according to Dexerto.

On that rig he ran a council of chatbots that voted on the best answer. Ajax is the next step, a model sized for ordinary home PCs rather than a 10-GPU tower.

If you want to try local AI yourself before Ajax ships, the graphics card matters most. Our guide to the best GPU for AI compares cards for running models like Qwen at home.

Learn the basics in our step-by-step guide on how to run an LLM locally. Then pick a machine from our list of the best desktop for AI.

Sources

TBreak, Free Press Journal, Dexerto, Odysseus launch, Dexerto, AI council. *Photos: PewDiePie at PAX 2015 by camknows, CC0, cropped and resized; Asus Strix RTX 4090 operational by Benlisquare, CC BY-SA 4.0, cropped and resized.*