PS5LM runs GGUF language models locally on a jailbroken PS5 using llama.cpp’s Vulkan backend. Its native-app route shares roughly 11.4 GB of memory between model and context and provides OpenAI- and Anthropic-compatible APIs. The developer reports about 55 tokens per second for a quantized Qwen 3.5 9B model on a PS5 Slim, but this is not independently verified. Documentation warns that GPU memory can remain allocated after model changes, requiring an app restart. This is not an app for an ordinary unmodified console.