Developer Turned PlayStation 5 Into AI Server
A new project enables jailbroken consoles to run language models using the system's high-bandwidth GPU.
Updated on Oct. 14, 2026 in Artificial Intelligence
Live Poll
Would you modify your home gaming console to run advanced AI models?
Developer cobanov has released the PS5LM project, which allows jailbroken PlayStation 5 consoles to function as local artificial intelligence servers. By porting the llama.cpp inference engine, the project leverages the console's custom hardware to run large language models.
Why it matters
The project demonstrates that consumer gaming hardware possesses the high memory bandwidth necessary to host local AI models, effectively repurposing dormant console capacity for large-scale computation.
The PlayStation 5 hardware delivers 448 GB/s of memory bandwidth via 16GB of GDDR6 memory, enabling the 27-billion parameter Qwen model to reach speeds of 20.3 tokens per second at a 32k context window.
The players
cobanov
A software developer who authored the PS5LM project for jailbroken console hardware.
Alibaba
A multinational technology conglomerate responsible for developing the Qwen series of large language models.
The details
The PS5LM project utilizes the relapse browser exploit to execute custom code on firmware versions 7.00 through 13.60. It functions by porting a RADV-based Vulkan driver—a software interface used for high-performance graphics—to the console's FreeBSD-based operating system, routing AI inference tasks directly through the AMD graphics processing unit.
Timeline
August 14: Alibaba released the Qwen3.8-27B model.
October 14, 2026: The project was documented and published.
The Tech Race
This development follows the trend of optimizing the llama.cpp engine to run on non-traditional hardware beyond standard desktop environments. It marks a departure from PC-centric AI deployment, utilizing high-bandwidth gaming consoles to challenge the current server-grade GPU bottleneck.
Users with jailbroken consoles on specific firmware can currently utilize these tools to run local models without external server requirements. The project is released under the GPL-3.0 license, allowing hobbyists to build upon the existing porting framework.
The takeaway
The project confirms that modern console hardware is capable of sophisticated inference tasks previously reserved for high-end workstations. Watch for future performance benchmarks as the developer optimizes the Vulkan driver integration for different model sizes.
Further reading
For more on the hardware acceleration of local models, see Artificial Intelligence.
Live Poll
Would you modify your home gaming console to run advanced AI models?
