PewDiePie is stepping into the AI race with a project that is less about building the biggest model and more about making AI personal. Felix Kjellberg has introduced Ajax, a compact language model designed to run locally through his self-hosted Odysseus workspace. The project has also drawn attention because Kjellberg says OpenAI suspended his account twice during development, with one displayed email citing “distillation.” Ajax is built on Alibaba’s Qwen3.5-9B and aims to handle everyday tasks while keeping processing closer to the user’s own hardware.
PewDiePie’s Ajax AI model focuses on local, private computing
Ajax takes a different path from the giant cloud-based models dominating the AI industry. Its foundation, Qwen3.5-9B, has around nine billion parameters, making it far smaller than many frontier systems.That size is part of the appeal. Kjellberg is designing Ajax for practical jobs such as web searches, browsing, emails and calendars rather than trying to compete on every possible AI benchmark.The model is intended to operate as an “always-on” assistant inside Odysseus. A local setup could give users more control over their data and reduce the need to send routine requests to an external server.However, Ajax is still far from a finished consumer product. The project had not published final hardware requirements, independent benchmark results or downloadable model weights when the announcement was made. Its website was still listing the model as “Coming soon.”
PewDiePie’s OpenAI controversy adds another layer to Ajax
The most unusual part of the project involves how Kjellberg says he developed Ajax. He wanted to use outputs from a larger AI system to help create training material for his smaller model, a process known as knowledge distillation.OpenAI allows distillation in some contexts, but its terms restrict using its output to develop competing models and prohibit certain forms of automated extraction. Kjellberg says his account was suspended twice while working on Ajax. One email he displayed reportedly cited “distillation” as the reason.OpenAI has not publicly provided a specific explanation for those account actions, so Kjellberg’s account remains the main source for those details.Ajax is also designed to behave differently from many mainstream assistants. Its creators describe its refusal behaviour as having been “ablated” to create a freer system. Kjellberg reportedly used the open-source Heretic tool for this process.Still, the project retains limits around harmful requests. Its safety claims have not undergone an independent assessment, so they should not be treated as proof that Ajax is safe.Kjellberg plans more reinforcement learning, quantisation and benchmarking before a later release. For now, Ajax represents something broader than a celebrity-built chatbot. It reflects a growing interest in smaller AI models that can run locally, offer greater control and reduce dependence on powerful cloud platforms.
