PS5LM project runs Qwen 3.8-27B on a jailbroken PS5 GPU at 21 tokens per second

A project called PS5LM runs Qwen 3.8-27B on a jailbroken PS5 using the console's GPU, according to a 14 October 2026 post by Andrew Curran on X. The model uses heavy compression to fit within the console's 16GB of memory and generates about 21 tokens per second.

Curran reports that smaller models run faster on the same hardware: Gemma 4 E4B reaches 61 tokens per second and Llama 3.2 3B reaches 124 tokens per second. The only source so far is this single post, and the report does not include independent testing or published technical details.