PewDiePie details DIY AI project that he claims rivaled ChatGPT on coding tests

PewDiePie has revealed he spent months fine-tuning his own AI model, claiming it temporarily outperformed ChatGPT on a coding benchmark.

In a new YouTube video, the creator explained that the project began as a personal challenge to better understand machine learning, not to build a model from scratch. Instead, he fine-tuned an existing large language model using custom datasets and coding-focused benchmarks aimed at improving performance for AI coding agents.

According to PewDiePie, the model initially scored just 8% on a coding benchmark before gradually improving through retraining and format adjustments. After introducing reasoning data and refining the dataset, he claimed one iteration reached 19.6%, briefly surpassing ChatGPT’s score at the time.

However, he later discovered benchmark contamination, meaning some of the training data overlapped with benchmark questions, which invalidated the result and forced him to restart the process.

Following retraining on a more coding-specific base model, PewDiePie said performance climbed significantly, eventually reaching 36% and later 39.1% after post-training adjustments.

The project wasn’t without setbacks. He detailed repeated system crashes, overheating issues, and even a failed GPU during training, describing his hardware setup as heavily modified to handle the computational load.

Despite the hurdles, PewDiePie emphasized that the experiment was primarily about learning through trial and error. He acknowledged that strong results on a single benchmark don’t necessarily translate to broader AI capability and said further testing would be needed before considering any public release.

He also noted that newer models, including Qwen 3, now score higher on the same benchmark, meaning continued development would be required to remain competitive.

For now, the YouTuber says the experiment stands as a deep dive into AI development — and proof that even creators outside traditional tech circles can explore the rapidly evolving field.