AI benchmarking Archives

Winsage

October 15, 2026

We test the RTX Spark Superchip in Microsoft’s Surface Laptop Ultra — Nvidia’s Windows on Arm platform impresses in local AI, but gaming isn’t its strong suit

Nvidia's CEO Jensen Huang introduced the RTX Spark platform at Computex 2026, which integrates the GB10 Superchip with a specialized version of Windows on Arm for local AI execution. Nvidia collaborates with Microsoft, Dell, HP, MSI, Lenovo, and Asus to create PCs that function as autonomous companions. The platform's local AI performance is assessed using a new benchmark called AgentPerf-local, focusing on completion time for AI tasks. The RTX Spark-powered Surface Laptop Ultra has shown good performance, though it comes with a high price of ,899 for 1TB of storage. The platform faces challenges due to rising DRAM and NAND prices and is still in development, with full software infrastructure yet to be established.

AppWizard

March 21, 2025

A high schooler built a website that lets you challenge AI models to a Minecraft build-off

Developers are exploring new methods to benchmark generative AI models, with one initiative being the Minecraft Benchmark (MC-Bench), a platform for head-to-head competitions among AI models that generate unique Minecraft creations. Users vote on the performances without knowing which AI created each entry. The project, created by 12th-grade student Adi Singh, leverages Minecraft's universal recognition to evaluate AI capabilities. MC-Bench currently has eight volunteer contributors and has received support from major AI companies like Anthropic, Google, OpenAI, and Alibaba. The focus is on simple builds, with plans to scale to more complex tasks. MC-Bench requires models to write code for requested builds, making it easier for users to assess the quality of creations visually. Singh believes the scores from MC-Bench provide meaningful insights into AI performance compared to traditional text-based benchmarks.