PS5LM runs llama.cpp and local LLMs on a jailbroken PlayStation 5 Other

PS5LM, an open-source project on GitHub, ports llama.cpp to the PlayStation 5, so the console runs local language models on its own CPU and GPU. Its author reports Qwen 3.8 27B answering at 9 to 21 tokens per second on the console's GPU. It needs a jailbroken PS5.

The clip shows the small Qwen3.5 0.8B model chatting in the PS5's browser on the console's CPU, at about 20 tokens per second.

What it is

• A port of llama.cpp, not a one-model hack. Earlier PS5 LLM work wrote GPU kernels for one model at a time. PS5LM brings llama.cpp itself, so any model and quantization llama.cpp supports can run: Qwen, Llama, Mistral, Gemma, from 1-bit to 8-bit GGUF files.
• Two ways to run it.

• Ways to talk to it. Type in the PS5's browser with the DualSense, open the model library from a phone or computer on the same network (port 8082), or point any client at the chat on port 8081, which speaks the OpenAI API. The GPU app also shows a dashboard on the TV: speed, GPU and memory use, and the loaded model.
• Models are plain GGUF files. Download them in the library, or copy them to the console over FTP or from a USB drive.

What you need

• A PS5 you own on firmware 7.00 to 13.60, jailbroken, with a payload loader.
• The project is GPL-3.0 and is not affiliated with Sony. The repository has no Sony SDK files, keys, firmware or model weights.

Price

If memory keeps getting more expensive, a console that already sits under the TV is one more place to run a small model at home.

Sources

• PS5LM on GitHub
• PS5LM project page and demo
• llama.cpp