PewDiePie Releases Uncensored Ajax Model Based on Qwen3.5-9B; Says OpenAI Suspended His Accounts Twice

Highlights

Ajax is a fine-tuned Qwen3.5-9B model that PewDiePie says runs on a home PC and removes the model's refusal safeguards to enable a freer user experience. He reports two account suspensions by OpenAI, with one email explicitly citing "distillation" as the reason. PewDiePie plans two more days of reinforcement learning and evaluation for the next Ajax release, and asks fans to contribute training data while stressing responsible use.

Sentiment Analysis

• 
  The overall sentiment is mixed: there is enthusiasm for decentralized, runnable-at-home AI and criticism of large AI providers. The tone balances excitement about technical achievement with concern about platform enforcement and data practices. The progress bar below reflects this mixed sentiment with moderate intensity.
  
  
  

  
  60%
  

  
  
  
  

Article Text

Prominent YouTuber PewDiePie has published an AI model called Ajax, which he says was fine-tuned from Alibaba's Qwen3.5-9B and can run on a typical home computer. The model is presented as an uncensored, private assistant intended to perform everyday tasks such as web search, email handling, and calendar management. According to a recent video, PewDiePie removed the model's refusal mechanisms using open-source tools to create a less constrained user experience, while advising users to act responsibly and avoid causing harm.

PewDiePie described experiencing two account suspensions from OpenAI during development. He shared an OpenAI email that attributed one suspension to "distillation," a process in which outputs or reasoning from one model are used to train another. He said the first suspension was later reversed, and the second happened after he used the model to generate seed data for further training. He questioned how the platform detected his activities and suggested that some data practices by large AI providers resemble distillation.

Ajax is part of a personal AI workspace called Odysseus, which PewDiePie introduced earlier in the year. The Ajax release is described as the first version, designed to act as a resident, autonomous agent and assistant. While Ajax’s creators removed refusal safeguards to expand capability, they also stated the model is "not designed to provide dangerous actionable instructions," following legal counsel. PewDiePie gave an example of nuclear weapons to illustrate uncertainty about the practical limits enforced by major providers.

The reported model size, represented by the 9B suffix in Qwen3.5-9B, indicates roughly nine billion parameters. Part of PewDiePie’s plan for Ajax includes two additional days of reinforcement learning and subsequent rounds of review, quantization, and evaluation. Quantization will reduce numerical precision and file size, making the model easier to run on consumer hardware. He also encouraged fans to donate training data to improve Ajax’s performance, though he noted few contributions at the time of the announcement.

PewDiePie emphasized a preference for smaller, task-specific models trained with appropriate harnesses—external code that calls tools and orchestrates tasks—over the speculative trend toward trillion-parameter models. He argued that well-designed small models can provide effective performance for targeted use cases. The release drew comparisons to other autonomous agent projects and raised questions about platform policy, the definition of permissible model training techniques, and the boundary between acceptable fine-tuning and prohibited actions.

OpenAI has not issued a public statement regarding PewDiePie's account suspensions. The episode highlights ongoing tensions in the AI ecosystem over how companies monitor and enforce usage policies, the role of distillation and dataset provenance, and the emergence of locally runnable models that challenge centralized control. While Ajax aims to offer greater freedom and privacy for users, it also prompts discussion about responsible deployment and the safeguards needed to prevent misuse.

In summary, Ajax represents a move toward more accessible, locally executable AI agents built from large open models, accompanied by controversy over enforcement actions and data practices. PewDiePie plans further training and evaluation for Ajax, urging responsible contributions and positioning smaller, specialized models as a viable alternative to monolithic, resource-intensive architectures.

Key Insights Table