Introducing Chan AI

Fine-tuned on Qwen3:4B with 16-bit LoRA.

Training took 4.5 days on an RTX 3090.

Dataset: Several hundred million tokens of text-only 4chan posts.

This is our first model that we trained. Instead of creating the most accurate and smartest model, we purposely aimed for the worst possible model. Only by starting at the bottom can we truly see how far we can climb. Feel free to explore its quirks and let us know what you discover!

We also created a new acronym(AI) not as in artificial intelligence but as artificial idiocy

Use the Try it now then change the model via the button on the left and select chan ai to experiment with the model!

Model Terms & Conditions

By interacting with this model you agree that you may ask any question or prompt that does not violate the following:

All other content is permitted. Use responsibly.