AI World Model Plays Pokémon Red on a Single RTX 3080 Ti
Developer stmonty trained a small AI world model on one RTX 3080 Ti to play Pokémon Red, successfully selecting Squirtle as a starter after fine-tuning.
A developer using the name stmonty has trained a small AI world model on a single RTX 3080 Ti gaming GPU to play Pokémon Red. The model, which has about 12.5 million parameters, was written up on Sept. 20.
Training used 42,382 grayscale frames collected across more than 1,000 short runs, all starting from a save in Professor Oak's lab. Instead of predicting the entire next screen, the model relies on compressed summaries of 192 numbers.
After fine-tuning, the model successfully selected Squirtle as a starter. In over 100 runs from the same save, 52 plans obtained a starter, compared with zero for random button presses and one for an untrained model.
The project's code is open source on GitHub as lePokeRed, with a CUDA GPU recommended. Tom's Hardware covered the project.
In March 2026, a paper detailed a single JEPA model with about 15 million parameters trainable on a single GPU. Yann LeCun split from Meta last year to build world models with his AMI Labs.
Quick answers
What hardware was used to train the Pokémon Red AI?
A single RTX 3080 Ti gaming GPU.
How many parameters does the model have?
About 12.5 million parameters.
Where can I find the code?
The project is open source on GitHub as lePokeRed, with a CUDA GPU recommended.