{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/23f92f47f790497b8e026b49f86dcaef\" frameborder=\"0\" width=\"1150\" height=\"862\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":862,"width":1150,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":862,"thumbnail_width":1150,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/23f92f47f790497b8e026b49f86dcaef-3da9bb8481d4c50b.gif","duration":205.236,"title":"Mintoken Cuts AI Token Costs with Caching","description":"This Loom explains how Mintoken reduces AI token cost by using a self-evolving cache and routing to the cheapest capable model. It checks cached responses for a proper answer first, and if none exists it queries the lowest-cost model such as open source or cloud options like ChatGPT or Gemini, then caches the result. The presenter claims users see up to 91% cost savings and that the system improves as more people use it. It also describes how tools like Pioneer, Senso, Guild, Ban, and Replay are used to route questions, enforce a 90% quality floor, promote only policies that preserve quality while saving cost, announce promotions, and validate reused answers."}