{"type":"post","id":"loco-local-preference-learning-cuts-reasoning-model-output-845762","slug":"loco-local-preference-learning-cuts-reasoning-model-output-845762","title":"LoCo: local preference learning cuts reasoning-model output by roughly 40-45% without losing accuracy","description":"LoCo is a local preference-learning method that trains reasoning models to stop deliberating when further steps are unnecessary.","retrieval_nugget":"LoCo is a local preference-learning method that trains reasoning models to stop deliberating when further steps are unnecessary. In a preprint, its authors report shorter completions on mathematical and code benchmarks while maintaining or improving accuracy, a potential route to lower inference costs for teams running their own models.","published_at":"2026-11-04","updated_at":"2026-11-04","record_date":"2026-11-04","date_kind":"published_at","topics":["agent-economics"],"entities":["LoCo","Xi’an Jiaotong University","Tencent WeChat AI","DeepSeek-R1-Distill-Qwen-7B","DeepSeek","Qwen"],"related_patterns":["agent-economics-moves-to-completed-work"],"novelty_label":"new_fact","what_changed":"LoCo introduces local preference learning, reporting 41.7%–45.4% shorter math outputs and 23.4%–36.3% shorter code outputs while maintaining or improving accuracy.","prior_refs":[],"source_url":"https://arxiv.org/abs/2611.04814","source_urls":["https://arxiv.org/abs/2611.04814"],"schema_version":"newruntime-agent-readable-v0.2","stable_id":"post:loco-local-preference-learning-cuts-reasoning-model-output-845762","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/loco-local-preference-learning-cuts-reasoning-model-output-845762/","markdown":"https://newruntime.com/posts/loco-local-preference-learning-cuts-reasoning-model-output-845762.md","json":"https://newruntime.com/posts/loco-local-preference-learning-cuts-reasoning-model-output-845762.json"},"markdown_url":"https://newruntime.com/posts/loco-local-preference-learning-cuts-reasoning-model-output-845762.md","next_reads":[{"type":"topic","path":"/topics/agent-economics/","url":"https://newruntime.com/topics/agent-economics/","title":"Agent Economics","reason":"Explore the agent-economics topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/agent-economics.md","json_url":"https://newruntime.com/topics/agent-economics.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"}]}