XenovaCom has introduced a new benchmark for testing the performance of machine learning (ML) models in browsers using WebGPU, showcasing significant speed improvements. The benchmark, named Transformers.js WebGPU Embedding Benchmark, allows users to measure how much WebGPU can speed up ML models running locally in their browsers. According to tests shared by users, WebGPU has enabled a 40x and 17x faster execution of ML models compared to previous technologies. Additionally, XenovaCom has updated its in-browser applications, such as depth estimation with Depth Anything and background removal using the RMBG-v1.4 model, to support WebGPU, achieving performance boosts up to 50x, reducing operation times to under 200ms. These advancements suggest a substantial enhancement in running complex ML models like chatbots and the full-sized ColBERT model directly in browsers that support WebGPU.