Recent advances in large language models (LLMs) test-time computing have introduced the capability to perform intermediate chain-of-thought (CoT) reasoning (thinking) before generating answers. While increasing the thinking budget yields smooth performance improvements at inference time, the relationship between LLM capability, query complexity, and optimal budget allocation remains poorly understood for achieving compute-optimal inference. To address this challenge, we utilize self-consistency, the agreement among multiple reasoning paths, as a proxy for thinking necessity. We first identify that lower self-consistency indicates when queries require extended thinking to reach correct answers. Building on this insight, we introduce Sonata (Self-Consistency-Guided Adapter for Thinking Allocation), a lightweight approach that adaptively allocates thinking budgets to optimize the performance-efficiency tradeoff. Sonata includes a adapter trained offline on a calibration dataset to predict self-consistency directly from last layer hidden representations during the query prefilling stage. This prediction then guides on-the-fly budget allocation before thinking. The adapter is general, transferrable across diverse tasks once trained, and introducing almost zero computational overhead during inference. Notably, Sonata is orthogonal to existing CoT compression methods, enabling further efficiency gains when managing thinking budgets across queries. Extensive experiments on multiple models (Qwen3-8B, GPT-OSS-120B, Qwen3-235B-A22B, Intern-S1-mini) and benchmarks (AIME24, AIME25, GSM8K, MATH500, GPQA) demonstrate that Sonata achieves 20% to 80% reduction in thinking tokens while maintaining the same accuracy, or up to 5% improvement in accuracy with same token cost.
†Work done while at Apple
‡The University of North Carolina at Chapel Hill