Tencent's LoCo Teaches Reasoning Models When to Stop Thinking

Tencent's WeChat AI and Xi'an Jiaotong University proposed LoCo, a local preference learning method that teaches reasoning models to stop deliberating once they have enough information, cutting math-task tokens by about 45% without losing accuracy.

2026-11-04 ~ 2026-11-04 · 2 related posts

• Tencent's LoCo teaches reasoning models when to stop, cutting output length without hurting accuracy — tencent · 2026-11-04
• LoCo: local preference learning cuts reasoning tokens by up to 45% without accuracy loss — pmttyji · 2026-11-04