Nice Fold or Hero Call: Learning Budget-Efficient Thinking for Adaptive Reasoning

TL;DR

This paper introduces BET, a framework for adaptive reasoning that optimizes test-time compute by learning when to solve, fold, or call, significantly reducing token usage while maintaining or improving performance.

Contribution

It proposes a novel investment-based approach to adaptive reasoning, combining behavioral cold-start with GRPO to improve efficiency and effectiveness across multiple benchmarks.

Findings

BET reduces reasoning tokens by ~55% on average.

BET improves overall performance while saving compute.

BET transfers zero-shot to new reasoning tasks with efficiency gains.

Abstract

Large reasoning models (LRMs) improve problem solving through extended reasoning, but often misallocate test-time compute. Existing efficiency methods reduce cost by compressing reasoning traces or conditioning budget on perceived difficulty, yet largely overlook solvability. As a result, they may spend large budgets on queries beyond the model's capability while compressing hard-but-solvable queries that require deeper reasoning. In this work, we formulate adaptive reasoning as a computational investment under uncertainty, where budget should follow the expected return of reasoning rather than perceived difficulty alone. To instantiate this principle, we propose Budget-Efficient Thinking (BET), a two-stage framework that combines behavioral cold-start with GRPO under an investment-cost-aware reward. By aligning solve-or-fold decisions with rollout-derived solvability, BET learns three behaviors: (1) short solve, answering easy queries concisely; (2) nice fold, abstaining early when continued reasoning has near-zero expected return; and (3) hero call, preserving sufficient compute for hard-but-solvable queries. Across seven benchmarks and three base models, BET reduces reasoning tokens by ~55% on average while achieving overall performance improvements, and transfers zero-shot from mathematical reasoning to scientific QA and logical reasoning with comparable efficiency gains.