Defici
← Defici Newstech-news

Qualcomm Snapdragon X Elite Gen 2 Posts 85 TOPS, Enabling Fully On-Device LLM Inference for Agentic Apps

By Defici Editorial · 27 Jul 2026

The 85 TOPS Milestone

Qualcomm's Snapdragon X Elite Gen 2 achieves 85 TOPS on its Neural Processing Unit -- a 70% improvement over the first-generation X Elite's 50 TOPS. The company benchmarked Llama 3.1 13B at 40 tokens per second on a power draw of 12 watts, which falls within the thermal envelope of a slim laptop chassis.

The practical implication is that AI agents running local inference models can operate continuously in the background of a laptop without triggering fan noise, battery drain alerts, or thermal throttling -- removing the friction that has limited consumer adoption of always-on AI assistants.

Agent Applications Targeting Gen 2

Several independent software vendors have pre-announced Snapdragon X Elite Gen 2-optimised builds of their agent platforms. Common use cases include background monitoring agents that track price changes on classified and e-commerce platforms, draft-composition agents for marketplace sellers that suggest listing copy improvements in real time, and local deal-matching agents that maintain a buyer preference profile and surface relevant new listings without a cloud round-trip.

Availability

Snapdragon X Elite Gen 2 is expected in commercial devices from Lenovo, HP, Samsung, and ASUS in Q4 2026. Microsoft has confirmed that Windows 12 will ship with a local model cache manager designed specifically for the Gen 2 NPU.

ShareXWhatsAppLinkedIn

Get Defici News in your inbox