Qualcomm launches GenieX to run LLMs on Windows laptops

Qualcomm has launched GenieX, a new SDK designed to enable the execution of large language models on Windows laptops equipped with Qualcomm chips. The tool supports running models via CPU, GPU, or NPU backends.

• Users report achieving 20 tokens per second with Gemma 4 26B A4B and 10 tokens per second with Qwen 3.6 27B MTP on the GPU.
• The SDK allows compatibility with llama.cpp by supporting any Q4_0 GGUF model across CPU, GPU, and NPU hardware.

This release aims to help Qualcomm catch up with other major chipmakers in providing robust software development kits for local AI inference.