{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/2a05bbcd13054ac6aa0ed7504789cd3b\" frameborder=\"0\" width=\"1366\" height=\"1024\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1024,"width":1366,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1024,"thumbnail_width":1366,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/2a05bbcd13054ac6aa0ed7504789cd3b-51ae1865459a0a22.gif","duration":285.562,"title":"Superbrain Beats Claude Code","description":"This Loom demonstrates how Superbrain works and compares its cost and token efficiency against cloud code. On first open, Tokenfold builds local checkpoints for the entire codebase so code stays on the user device, then answers questions like “What is this project about?” using the selected model (SONET 4.6) in concise mode. For Superbrain, the cost is about 6 with roughly 32,000 read tokens, 10.7K write tokens, and 1.6K output tokens. For cloud code, the same question uses a “medium effort” setting and is reported to cost about 0.23 cents with around 150 read and 13,30.9K write tokens, producing 813 output tokens."}