When2Think: Learning Difficulty-Aware Length Control for Efficient Hybrid Reasoning Models

Hugging Face Daily Papers Papers

Summary

When2Think is a post-training framework that dynamically allocates computation in large reasoning models based on problem difficulty, improving accuracy-efficiency trade-offs on mathematical benchmarks.

Large Reasoning Models (LRMs) achieve strong performance on complex tasks but exhibit systematic inefficiency: they often overthink easy problems and underthink hard ones. Existing approaches based on uniform length penalties or rigid routing incur an efficiency tax, trading reduced computation on easy instances for accuracy loss on hard instances. We formulate efficient reasoning as an instance-adaptive computation allocation problem and propose When2Think, a post-training framework for hybrid reasoning that dynamically allocates computation based on problem difficulty. Our method introduces Instance-level Difficulty-Aware Control (IDAC), a reward-shaping mechanism that leverages pre-computed reference statistics (accuracy and token usage) to regulate reasoning depth. Combined with verifier-based rewards and batch-wise standardized advantages, IDAC enables stable critic-free optimization without learned reward models or online reference-model queries. When2Think encourages direct answering on easy instances while preserving extended reasoning on hard instances, thereby learning when to use System 1 (NoThink) versus System 2 (Think). Experiments on mathematical benchmarks demonstrate improved accuracy-efficiency trade-offs: on AIME24, Pass@3 increases by 10.0% while token usage is reduced by 27.9% relative to the base model, and on AIME25, When2Think achieves 40.0% Pass@3, outperforming compression and routing-only baselines.
Original Article
View Cached Full Text

Cached at: 09/18/26, 02:58 AM

Paper page - When2Think: Learning Difficulty-Aware Length Control for Efficient Hybrid Reasoning Models

Source: https://huggingface.co/papers/2609.19671

Abstract

LargeReasoningModels(LRMs)achievestrongperformanceoncomplextasksbutexhibitsystematicinefficiency:theyoftenoverthinkeasyproblemsandunderthinkhardones.Existingapproachesbasedonuniformlengthpenaltiesorrigidroutingincuranefficiencytax,tradingreducedcomputationoneasyinstancesforaccuracylossonhardinstances.Weformulateefficientreasoningasaninstance-adaptivecomputationallocationproblemandproposeWhen2Think,apost-trainingframeworkforhybridreasoningthatdynamicallyallocatescomputationbasedonproblemdifficulty.OurmethodintroducesInstance-levelDifficulty-AwareControl(IDAC),areward-shapingmechanismthatleveragespre-computedreferencestatistics(accuracyandtokenusage)toregulatereasoningdepth.Combinedwithverifier-basedrewardsandbatch-wisestandardizedadvantages,IDACenablesstablecritic-freeoptimizationwithoutlearnedrewardmodelsoronlinereference-modelqueries.When2Thinkencouragesdirectansweringoneasyinstanceswhilepreservingextendedreasoningonhardinstances,therebylearningwhentouseSystem1(NoThink)versusSystem2(Think).Experimentsonmathematicalbenchmarksdemonstrateimprovedaccuracy-efficiencytrade-offs:onAIME24,[email protected]%whiletokenusageisreducedby27.9%relativetothebasemodel,andonAIME25,When2Thinkachieves40.0%Pass@3,outperformingcompressionandrouting-onlybaselines.

View arXiv pageView PDFAdd to collection

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2609.19671 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2609.19671 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.19671 in a Space README.md to link it from this page.

Collections including this paper1

Similar Articles

DyCon: Dynamic Reasoning Control via Evolving Difficulty Modeling

arXiv cs.AI

This paper introduces DyCon, a training-free framework that uses step-level embeddings to model evolving task difficulty and dynamically control reasoning depth in Large Reasoning Models, effectively reducing overthinking and improving efficiency without sacrificing accuracy.