<?xml version="1.0" encoding="UTF-8"?><oembed><type>video</type><version>1.0</version><html>&lt;iframe src=&quot;https://www.loom.com/embed/7744d333b75a42a79fbcc672f69e5c67&quot; frameborder=&quot;0&quot; width=&quot;2446&quot; height=&quot;1834&quot; webkitallowfullscreen mozallowfullscreen allowfullscreen&gt;&lt;/iframe&gt;</html><height>1834</height><width>2446</width><provider_name>Loom</provider_name><provider_url>https://www.loom.com</provider_url><thumbnail_height>1834</thumbnail_height><thumbnail_width>2446</thumbnail_width><thumbnail_url>https://cdn.loom.com/sessions/thumbnails/7744d333b75a42a79fbcc672f69e5c67-e9e47f9e9a023d61.gif</thumbnail_url><duration>138.941</duration><title>Benchmarking model decode rates with Antigravity 🚀</title><description>Hey Alex, I wanted to show off a project I did this weekend using Antigravity instead of Cloud Code for the first time. I built a UI that benchmarks multiple models and logs results like time to first token, prompt processing versus decode, and tokens per second, with averages saved into history after prompt processing finishes. I store the runs in a SQLite database and the project is in Go and Mantean, with requirements in the public README. I also tried serving it on GCP and hit an out of memory error to debug later. Let me know if you have any questions.</description></oembed>