← All stories
● Covered by 1 source · 1 reportMedium impact1 neutral

Jailbroken PS4 runs 9-billion-parameter AI model offline at up to 43 tokens/second

🔄 Updated 5d ago
New to BrevFeed? We gather this story from every outlet covering it into one summary — ranked by real-world impact, not just the latest headline — so you never miss what matters. What is BrevFeed? →

Key points

  • Garlic LLM runs Qwen3.5-9B on a jailbroken PS4.
  • Model operates offline at 31-39 tokens/second for text chat.
  • Supports voice conversation and dictation in multiple languages.
  • Uses OpenCL C kernels for GPU operations, direct GPU access.

Garlic LLM Enables Offline AI on PS4

Developer Leonardo Alegría has released Garlic LLM, an open-source project that allows a jailbroken PlayStation 4 to run the 9-billion-parameter Qwen3.5-9B AI model offline. The application utilizes the console's 2013-era GPU, achieving speeds of 31 to 39 tokens per second for text chat. The model is compressed to approximately 4.5GiB to fit within the PS4's memory.

Technical Implementation and Availability

Garlic LLM operates on PS4 consoles running GoldHEN, a homebrew enabler. The application directly communicates with the GPU, bypassing standard driver layers. It is available as a free 5.66GB download on Hugging Face and can be installed via the PS4’s Package Installer, requiring about 6GB of free space. The developer tested the project on base/Slim PS4 models with firmware 11.52.

The model's operations, such as matrix multiplications, are handled by compute kernels written in OpenCL C and compiled with LLVM for AMD's Graphics Core Next (GCN) architecture. The CPU manages the tokenizer, interface, text rendering, and audio components.

Voice Capabilities and Language Support

The project includes voice conversation capabilities, offering spoken replies in seven languages through the Chatterbox speech model, with the chat model running at approximately 22 tokens per second. Users can also utilize a held controller button for dictation, powered by Nvidia’s Parakeet TDT 0.6B v3 speech-to-text model, which supports 25 European languages.

Comparison to Other Console AI Projects

This project differs from efforts like the PS5LM fork, which runs llama.cpp as a native application on the PlayStation 5, accessing its AMD GPU through PS5_Vulkan, a port of the RADV driver. Garlic LLM does not use llama.cpp or a Vulkan port, instead building its own low-level PM4 command buffers for direct GPU interaction on the PS4.

✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →

The daily brief

One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.

One email a day. Unsubscribe in one click, any time.

Today's brief

Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.

~4 min · 3 stories · Nov 21

▶ Play today's brief Listen on Spotify

New every morning, and the back catalogue is archived by date.

Reporting from

Developer Leonardo Alegría's open-source project, Garlic LLM, enables a jailbroken PlayStation 4 to run the Qwen3.5-9B AI model offline on its 2013-era GPU. This allows the console to process AI tasks, including voice conversation and dictation, at speeds up to 43 tokens per second, by directly accessing the GPU and fitting the model into the console's memory.