Tencent's LoCo teaches reasoning models when to stop, cutting output length without hurting accuracy
tencent · hf · 2026-11-04
A Tencent (WeChat AI) and Xi'an Jiaotong University paper introduces LoCo, a local preference learning method for reasoning compression that targets overthinking — models deliberating past the point of sufficient information.
- Unlike methods operating on full reasoning trajectories, LoCo focuses on the local decision of whether further reasoning is needed: it detects candidate overthinking states in the model's own traces, freezes the reasoning prefixes, and regenerates shorter verified continuations from the same states.
- These state-conditioned preference pairs train with DPO, teaching the model when to continue versus conclude rather than just favoring shorter outputs.
- Training data is built entirely from self-generated math traces; across math and code tasks, output length drops substantially while accuracy holds or improves.
More from Research
- Google and Mercor's PLE benchmark: all frontier agents score below 50% on 85 real engineering tasks — teortaxesTex · 2026-11-04
- Fable 5.1 hand-writes 100k-param classifier weights, beating backprop training — AccBalanced · 2026-11-04
- Cambridge's van der Schaar Lab opens PhD applications for 2027 entry — MihaelaVDS · 2026-11-04
- Tencent's Hyra 2.0 Generates Protein Sequences 57x Faster Than ProteinMPNN — YvesMulkers · 2026-11-04
- Researcher Urges Top AI Conferences to Desk-Reject AI-Generated Papers Amid Fake Data Scandals — Vegetable-Formal-753 · 2026-11-04
- Harvard Professors Admit AI Article Reviews Now Beat the Best Humans — iamfakhrealam · 2026-11-04