Batched Contextual Reinforcement: A Task-Scaling Law for Efficient Reasoning
2026-04-02 • Machine Learning
Machine LearningArtificial IntelligenceComputation and Language
AI summaryⓘ
The authors propose a new training method called Batched Contextual Reinforcement (BCR) that teaches large language models to solve multiple problems at once in a shared space. This approach reduces the number of tokens the model uses without hurting accuracy, unlike older methods that trade off quality for efficiency. They found that increasing the number of problems solved together decreases token usage and only slightly affects accuracy. Their method also avoids problems seen with previous techniques, like unstable training. Overall, the authors show that a simple change in training setup can make reasoning models much more efficient.
Large Language ModelsChain-of-ThoughtToken EfficiencyReinforcement LearningInference CostContext WindowTraining ParadigmLength PenaltyMathematical BenchmarksOptimization Stability
Authors
Bangji Yang, Hongbo Ma, Jiajun Fan, Ge Liu
Abstract
Large Language Models employing Chain-of-Thought reasoning achieve strong performance but suffer from excessive token consumption that inflates inference costs. Existing efficiency methods such as explicit length penalties, difficulty estimators, or multi-stage curricula either degrade reasoning quality or require complex training pipelines. We introduce Batched Contextual Reinforcement, a minimalist, single-stage training paradigm that unlocks efficient reasoning through a simple structural modification: training the model to solve N problems simultaneously within a shared context window, rewarded purely by per-instance accuracy. This formulation creates an implicit token budget that yields several key findings: (1) We identify a novel task-scaling law: as the number of concurrent problems N increases during inference, per-problem token usage decreases monotonically while accuracy degrades far more gracefully than baselines, establishing N as a controllable throughput dimension. (2) BCR challenges the traditional accuracy-efficiency trade-off by demonstrating a "free lunch" phenomenon at standard single-problem inference. Across both 1.5B and 4B model families, BCR reduces token usage by 15.8% to 62.6% while consistently maintaining or improving accuracy across five major mathematical benchmarks. (3) Qualitative analyses reveal emergent self-regulated efficiency, where models autonomously eliminate redundant metacognitive loops without explicit length supervision. (4) Crucially, we empirically demonstrate that implicit budget constraints successfully circumvent the adversarial gradients and catastrophic optimization collapse inherent to explicit length penalties, offering a highly stable, constraint-based alternative for length control. These results prove BCR practical, showing simple structural incentives unlock latent high-density reasoning in LLMs.