Back to Tool
P

PS5LM

Tool Oct 16, 2026

A community llama.cpp port that serves local language and decision models from a jailbroken PS5's GPU.

Visit primary link 3 sources attached
PS5LM's author-published TV interface showing Qwen 3.8 27B, a local network endpoint and streaming requests.
Screenshot included in PS5LM's v0.9.0 tagged README. The screen itself displays version 0.8.3; its 20.7 tokens-per-second reading is the author's displayed output, not a reproduced AITrending benchmark. Full screen preserved. Source: Mert Cobanov / PS5LM contributors
self-hostedagent-ecosystemdecision-models

What happened lately

PS5LM timeline

PS5LM runs llama.cpp through Vulkan and RADV on a jailbroken PlayStation 5. Its native app manages GGUF models and serves clients on the local network. Version 0.9.0 adds typed decision-model requests and a TV interface. It is GPL-3.0-or-later; llama.cpp retains its MIT license. The project is not affiliated with Sony.