---
id: "2026-10-16-ps5lm-llama-cpp-jailbroken-ps5"
url: "https://postcutoff.com/e/2026-10-16-ps5lm-llama-cpp-jailbroken-ps5/"
as_of: "2026-10-17T22:10:00+02:00"
date: "2026-10-16"
date_precision: day
category: open-source
importance: 2
confidence: high
status: [Confirmed]
sources: 2
editor: human editor
human_review: null
version: null
---

As of: 2026-10-17 22:10 CEST.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/e/2026-10-16-ps5lm-llama-cpp-jailbroken-ps5/

# PS5LM runs local LLMs on a jailbroken PS5

Full title: PS5LM: llama.cpp port runs GGUF models on a jailbroken PS5's GPU and serves them over OpenAI- and Anthropic-compatible APIs

An open-source project, PS5LM (GitHub user cobanov, GPL-3.0), ports llama.cpp to the GPU of a PlayStation 5 jailbroken on firmware 7.00–13.60 and serves GGUF models on the home network through OpenAI- and Anthropic-compatible APIs. It lists ~55 tokens/s for a 9B model and ~20 tokens/s for a 27B model at 2-bit on a PS5 Slim; it reached 360 points on Hacker News.

## Key facts
- Requirements: PS5 on firmware 7.00–13.60, jailbroken, with the kstuff-lite, etaHEN and ShadowMountPlus payloads; model and context share about 11.4 GB of console memory
- Listed speeds (PS5 Slim): Qwen 3.5 0.8B Q4_K_M 207 tok/s; Qwen 3.5 9B Q4_K_M 55 tok/s; Qwen 3.8 27B UD-Q2_K_XL about 20 tok/s with 64k context
- Endpoints: OpenAI-compatible and Anthropic-compatible APIs, plus a /v1/systemone endpoint for small decision models
- Reach: Hacker News front page, 360 points (Oct 16, 2026); the repository gives no exact release date or AI-assistance disclosure

## What happened

After the PS5 jailbreak for firmware up to 13.60, a hobbyist ported llama.cpp so the console's GPU can run quantised open-weight models
and act as a local inference server for coding tools that speak the OpenAI or Anthropic API.

## Why it matters

It shows how far local inference has spread: a games console with about 16 GB of shared memory can serve a 27B model at usable speed.

## Your AI and this story

- GPT-6 Astra (training cutoff April 2026): 169 days after its cutoff
- Claude Opus 5.5 (training cutoff June 2026): 108 days after its cutoff
- Gemini 4 Argon (training cutoff March 2026): 199 days after its cutoff
- Grok 4.7 (training cutoff May 2026): 138 days after its cutoff

## Sources

1. [GitHub: cobanov/PS5LM](https://github.com/cobanov/PS5LM) (github.com, code)
2. [Hacker News discussion](https://news.ycombinator.com/item?id=50122431) (news.ycombinator.com, discussion)

## Changes

- 2026-10-17 (filed): Created (read the GitHub README)

## Related

- 2026-10-08: [Hundreds of AI-decompiled game ports appear](https://postcutoff.com/e/2026-10-08-ai-decompiled-browser-game-ports/index.md)