YouTuber PewDiePie claims OpenAI banned his account twice while he was developing Ajax, a small AI model designed to run locally on users’ computers and browse the web.
PewDiePie discussed the experience in a new video, explaining that Ajax is a fine-tuned model built for Odysseus, a free, self-hosted AI application he launched in June. Running AI locally can keep data such as emails and calendars on a user’s own computer, while avoiding subscription fees.
Ajax is reportedly a nine-billion-parameter model based on Alibaba’s open-source Qwen 3.5. PewDiePie said the model is still being developed and will be released on his website once he considers it ready.
According to PewDiePie, he initially wanted to distil knowledge from OpenAI’s GPT-5.6 Sol to create training data for Ajax. He claimed his account was banned after he attempted to obtain the model’s encrypted reasoning tokens, which OpenAI keeps hidden.
PewDiePie said his account was later restored after he disputed the ban. However, he claimed OpenAI banned him again after he used the model to generate seed data for Ajax.
OpenAI’s terms prohibit using its outputs to develop competing models, meaning such activity could violate the company’s usage rules.
PewDiePie also said he used an open-source tool called Heretic to remove some of Ajax’s built-in refusals, making the model less restrictive. He described the process as giving the model “a little brain damage”, saying it currently succeeds at tasks such as organising inboxes and browsing the web around nine times out of 10.
He said Ajax is still being trained using GRPO, a method that allows the model to learn by attempting the same task multiple times and improving based on successful results.
PewDiePie said the project is aimed at demonstrating what smaller, locally run AI models can achieve without relying on large cloud-based systems. However, Ajax is not yet available for download as he continues testing, quantising and benchmarking the model.
