{"type":"post","id":"modelbest-lookahead-expert-scheduling-les-running-a-35b-3645b3","slug":"modelbest-lookahead-expert-scheduling-les-running-a-35b-3645b3","title":"ModelBest Lookahead Expert Scheduling (LES): Running a 35B MoE On-Device in Real Time, Losslessly","description":"ModelBest says Lookahead Expert Scheduling (LES) lets MiniCPM-V 4.7, a 35-billion-parameter MoE model, decode above 20 tokens per second on an iPhone 18 Pro Max without changing the model's routing or network structure.","retrieval_nugget":"ModelBest says Lookahead Expert Scheduling (LES) lets MiniCPM-V 4.7, a 35-billion-parameter MoE model, decode above 20 tokens per second on an iPhone 18 Pro Max without changing the model's routing or network structure. The result matters to builders because it shows how storage, scheduling, quantization, and speculative decoding can make a large model usable on constrained devices.","published_at":"2026-11-08","updated_at":"2026-11-08","record_date":"2026-11-08","date_kind":"published_at","topics":["local-models","inference","models"],"entities":["ModelBest","OpenBMB","Lookahead Expert Scheduling","MiniCPM-V 4.7","DSpark","llama.cpp"],"related_patterns":[],"novelty_label":"new_fact","what_changed":"ModelBest details LES, a lossless scheduling system that runs MiniCPM-V 4.7’s 35B MoE model at over 20 tokens per second on an iPhone 18 Pro Max.","prior_refs":["https://newruntime.com/posts/modelbest-and-openbmb-launch-minicpm-v-4-7-23d8e1/"],"source_url":"https://openbmb.cn/en/blog/les-35b-moe-on-device-real-time","source_urls":["https://openbmb.cn/en/blog/les-35b-moe-on-device-real-time"],"schema_version":"newruntime-agent-readable-v0.2","stable_id":"post:modelbest-lookahead-expert-scheduling-les-running-a-35b-3645b3","status":"published","visuals":[],"editor":{"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]},"routes":{"html":"https://newruntime.com/posts/modelbest-lookahead-expert-scheduling-les-running-a-35b-3645b3/","markdown":"https://newruntime.com/posts/modelbest-lookahead-expert-scheduling-les-running-a-35b-3645b3.md","json":"https://newruntime.com/posts/modelbest-lookahead-expert-scheduling-les-running-a-35b-3645b3.json"},"markdown_url":"https://newruntime.com/posts/modelbest-lookahead-expert-scheduling-les-running-a-35b-3645b3.md","next_reads":[{"type":"topic","path":"/topics/local-models/","url":"https://newruntime.com/topics/local-models/","title":"Local Models","reason":"Explore the local-models topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/local-models.md","json_url":"https://newruntime.com/topics/local-models.json"},{"type":"topic","path":"/topics/inference/","url":"https://newruntime.com/topics/inference/","title":"Inference","reason":"Explore the inference topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/inference.md","json_url":"https://newruntime.com/topics/inference.json"},{"type":"topic","path":"/topics/models/","url":"https://newruntime.com/topics/models/","title":"Models","reason":"Explore the models topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/models.md","json_url":"https://newruntime.com/topics/models.json"},{"type":"collection","path":"/posts/","url":"https://newruntime.com/posts/","title":"Field Notes","reason":"Open the latest editorial analysis.","media_type":"text/html","json_url":"https://newruntime.com/posts.json"},{"type":"collection","path":"/sources/","url":"https://newruntime.com/sources/","title":"Sources","reason":"See the original sources this archive cites.","media_type":"text/html"}]}