Developer abhi245y has released PS5LM Plus, an application that enables the execution of large language models on jailbroken PlayStation 5 consoles. The project builds upon the original PS5LM by AI engineer Mert Cobanov, shifting the computational workload from the CPU to the console's GPU.
To optimize performance, the developer ported the AMD RADV driver from Mesa and adapted llama.cpp to utilize Vulkan. These adjustments increased the available memory for models from 6 GB to 11.4 GB. With a memory bandwidth of 448 GB/s, the Qwen3.8-27B model achieved a generation speed of 20.9 tokens per second.
During testing, the GPU load reached 100 percent, while CPU usage remained at 2 percent. Smaller models demonstrate higher performance: the Gemma 4 E4B model generates approximately 61 tokens per second, and Llama 3.2 3B reaches up to 124 tokens per second.
The software includes a web interface and provides an API via port 8081 for local network access. It supports OpenAI and Anthropic interfaces, allowing integration with tools such as OpenCode, Codex, and Claude Code. Users are limited by the console's 16 GB of RAM and the requirement for a jailbroken system, which carries risks of account or hardware bans.
