Jailbroken PS5 Runs a 27B AI Model at Over 20 Tokens Per Second

Notably, a jailbroken PlayStation 5 is now being applied to run a 27-billion-parameter AI model directly on the console's GPU, reaching generation speeds of around 20 tokens per second.

Key points

• While a longer 7, 775-token response averaged roughly 19.7 tokens per second, another 9.2GB quantized version reached around 20.3 tokens per second.
• Qwen 3.8 27B Runs at Around 20 Tokens Per Second.
• A project called PS5LM has managed to run Qwen 3.8 27B, an AI model with around 27 billion parameters, using the PS5's GPU.
• PS5LM Plus has successfully run Qwen 3.8 27B, an AI model with roughly 27 billion parameters.
• A 6.8GB IQ2_XXS version of the model reached around 20.7 tokens per second with a 64K context window.

In practice, the experiment comes from PS5LM Plus, a fork of Mert Cobanov's PS5LM project. While PS5LM Plus adds a native application that can employ the console's GPU, the original project brought llama.cpp to a jailbroken PS5 through a CPU-based payload.

Notably, the project additionally gained attention after @Pirat_Nation posted concerning it on X, highlighting that a jailbroken PS5 was running Qwen 3.8 27B locally at around 20 tokens per second. PlayStation Controllers Could Shortly Secure Tap-to-Pay Feature. Qwen 3.8 27B Runs at Around 20 Tokens Per Second.

For context, the PS5 has 16GB of shared GDDR6 memory, so the model requires heavy compression to fit into the available memory.

For context, the test applied a heavily quantized version of Qwen 3.8 27B, with the application able to access around 11.44GB of GPU and CPU shared memory.

In practice, a jailbroken PS5 is now being applied to run a powerful AI model directly on the console.

Meanwhile, a project called PS5LM has managed to run Qwen 3.8 27B, an AI model with around 27 billion parameters, using the PS5's GPU.

In practice, the PS5 has only 16GB of memory, so the AI model needs heavy. Pic.twitter.com/hZCu8c7zLs. Pirat_Nation 🔴 (@Pirat_Nation) October 14, 2026.

In practice, a 6.8GB IQ2_XXS version of the model reached around 20.7 tokens per second with a 64K context window.

Notably, the project dashboard indicated generation at around 20.9 tokens per second. GPU Hits 100% While CPU Usage Stays Low.

While CPU usage remained particularly low, during testing, the PS5's GPU was running at 100% utilization.

For context, the console consumed around 86W of power and reached temperatures of roughly 54°C to 55°C.

This demonstrates that most of the AI workload was being handled by the PS5's GPU rather than its CPU. Smaller Models Run Much Faster.

Smaller AI models achieved significantly higher speeds on the same console. While Llama 3.2 3B reached as high as 124 tokens per second, gemma 4 E4B reached around 61 tokens per second.

For context, the difference between CPU and GPU execution is particularly sizeable. Qwen 3.8 27B reportedly managed only around one token per minute when streamed from storage and run through the earlier CPU-based PS5LM setup. PS5 and Xbox Sales Are Crashing – Here's Why. PS5 Can Act as a Local AI Server.

PS5LM Plus runs llama.cpp directly on the PS5 GPU and includes a dashboard showing generation speed, GPU and CPU usage, memory consumption, power draw, and temperatures.

It can load GGUF models from internal storage, USB storage, or an installed M.2 drive.

For context, the project can additionally expose an OpenAI-compatible API over the local network, allowing another computer or phone to interact with the AI model while inference runs on the PS5.

PS5LM Plus is a community homebrew project rather than an official Sony feature and requires a jailbroken PlayStation 5. Stay Connected with ProPakistani.

Obtain the latest tech news, telecom insights, and product launches wherever you prefer. Follow on Google Discover. Follow on Google News Join WhatsApp. See more ProPakistani stories in Google Search and Top Stories.

Technology and Automotive Specialist covering the latest cars, smartphones, AI breakthroughs, and.

In short, jailbroken PS5 Runs a 27B AI Model at Over 20 Tokens Per is the central thread here, and readers can expect follow-up reporting as the picture becomes clearer.