TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
PS5LM is a homebrew app that runs llama.cpp on a jailbroken PlayStation 5 and serves loaded GGUF models to devices on the same network. Its developers report performance figures for selected models on a PS5 Slim, but the project requires specific firmware, a jailbreak and supporting payloads.
The PS5LM homebrew project says its app can run GGUF models locally on a jailbroken PlayStation 5 using llama.cpp and the console’s GPU, then serve model responses to other devices on a home network. The project describes support for OpenAI-compatible and Anthropic-compatible interfaces, but installation requires a PS5 on firmware 7.00 to 13.60 with a jailbreak and additional payloads.
According to the project’s GitHub documentation, PS5LM runs llama.cpp with its Vulkan backend on the PS5’s RDNA 2 GPU. It allocates up to 11.4 GB of shared memory for a model and its context, with the amount available depending on the console’s free memory and the model loaded. The app can download models from Hugging Face or load GGUF files copied to console storage or a USB drive.
PS5LM serves requests on the local network at http://<console>:8081. The project lists compatibility with the OpenAI /v1/chat/completions endpoint and Anthropic’s /v1/messages endpoint, alongside a browser-based chat page. It says requests and generated replies can be viewed on the TV interface. The app also includes console monitoring and a planner that selects context and cache settings based on available memory.
The project’s performance table reports that, on a PS5 Slim, Qwen 3.5 0.8B at Q4_K_M generated 207 tokens per second, while Qwen 3.5 9B at Q4_K_M generated 55 tokens per second. It lists a 27B model at about 20 tokens per second with a 64,000-token context. These are developer-reported results for selected configurations; the supplied material does not describe an independent test or a standardised comparison.
A Console as a Local Model Server
PS5LM repurposes console hardware as a local inference server, so a laptop, phone or compatible coding tool can send requests to a model running on the PS5. The project says the setup can run without a separate PC, cloud service or account while serving requests on the home network. For users who already own a supported, jailbroken console, that offers another way to experiment with open models and local AI interfaces.
The practical reach is limited by the need for a specific firmware range and jailbreak, plus the technical work of installing payloads and transferring the app. The published speed figures also vary by model size and quantisation, so the larger-model results should not be read as evidence that every GGUF model will fit or run at a useful rate. PS5LM’s claim of support for architectures understood by llama.cpp is bounded by the runtime’s own model support and the console’s available memory.
The project also adds a decision-model mode that returns structured answers rather than generating a passage of text. Its documentation says these models can answer typed questions about text and, for models with an image projector, images. That broadens the app’s use beyond chat, though the project’s examples and measurements are not an independent evaluation of accuracy.
external USB drive for gaming consoles
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Firmware, Payloads and Model Support
PS5LM is presented as a native homebrew app, not an official Sony feature. Its documentation credits earlier PS5 GPU work and other homebrew projects, and says the app is built with llama.cpp and Vulkan-related components. The repository states that it is not affiliated with Sony Interactive Entertainment.
Installation requires a PS5 running firmware 7.00 through 13.60 and a jailbreak. The project lists kstuff-lite and etaHEN as required payloads, with ShadowMountPlus used to place the app on the home screen. It recommends additional tools for temperature, power and storage readings, and for copying files. Users can load a suggested model through the app or place a GGUF file in its model folder.
The source also describes compatibility with decision models carrying llama.cpp decision metadata, including named examples such as d1 Omni and Clef Flash. It reports that models with image projectors can accept images. Those capabilities depend on the particular model file and its compatibility; the source does not establish that every GGUF model supports image input or decision-style output.
““PS5LM is for anyone with a jailbroken PS5 who wants a private model server at home.””
— PS5LM project documentation
As an affiliate, we earn on qualifying purchases.
Compatibility and Results Still Vary
The supplied project material does not identify the release date or provide independent verification of its performance table. It does not specify a testing protocol, repeated-run results, or how performance changes across PS5 models, system conditions and model settings. The listed token rates should be treated as project-reported measurements, not guaranteed results.
It is also unclear how much memory will be available on a particular console after system use, and which models will fit with useful context lengths. The documentation’s broad compatibility language refers to architectures and quantisations supported by llama.cpp, but does not mean every GGUF file will load successfully. The source gives no evidence that the app is approved or supported by Sony, and jailbreaking is a prerequisite rather than an official installation path.
As an affiliate, we earn on qualifying purchases.
Installation and Project Updates
Users who meet the stated firmware and jailbreak requirements can install the app from a PS5LM release, add its supporting payloads, and load a suggested model or copy a GGUF file to the console. The project directs users to its documentation for building, console notes, model recommendations and command-line client setup.
The repository lists issues, a roadmap and a changelog, but the supplied material does not set out a dated next release or a planned compatibility expansion. Further project updates would clarify which models and console configurations have been tested, and whether the reported performance figures hold across a wider set of devices and workloads.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is PS5LM?
PS5LM is a homebrew app that runs llama.cpp on a jailbroken PS5 and makes a loaded model available to devices on the same network.
Can it run any GGUF model?
The project says it supports models and quantisations that llama.cpp can handle. That does not guarantee that every GGUF file will load or fit in the console’s available memory.
Which PS5 firmware versions are listed?
The documentation gives a supported range of firmware 7.00 to 13.60 and requires a jailbreak, along with additional homebrew payloads.
How fast does it run models?
PS5LM reports, for example, 55 tokens per second for Qwen 3.5 9B at Q4_K_M on a PS5 Slim. These are project-reported figures, and the source does not provide independent testing details.
Can other devices use the model?
Yes, according to the project. It serves requests on the local network at port 8081 and lists OpenAI- and Anthropic-compatible API endpoints, as well as browser chat.
Source: hn
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
