MLCommons releases MLPerf Client v2.0 for AI PC benchmarking

MLCommons has released MLPerf Client v2.0, an updated industry-standard benchmark designed to evaluate artificial intelligence performance on personal computers, including laptops, desktops, and workstations.

The new version expands testing capabilities beyond traditional Large Language Model (LLM) inference to include new categories for image generation and agentic AI. The image generation category utilizes Flux. 2 klein 4B as an experimental test, while agentic AI performance is measured through Software Engineering (SWE) Agent and Data Analyst Agent scenarios. These updates aim to capture the growing demand for autonomous software agents and visual generative capabilities.

Additionally, the benchmark has updated its LLM inference tests, incorporating Phi 4 Mini Instruct and introducing Qwen 3 8B as an experimental model. The tool continues to report metrics for both responsiveness and throughput to provide a comprehensive view of how local hardware handles evolving AI workloads.