Language Models can Self-Improve at State-Value Estimation for Better Search

Collecting ground-truth rewards or human demonstrations for multi-step reasoning tasks is often prohibitively expensive, particularly in interactive domains such as web tasks. We introduce Self-Taught Lookahead (STL), a reward-free framework that improves language model-based value functions by reasoning explicitly about state transitions. STL can be viewed as a chain-of-thought analogue of the value iteration algorithm: instead of regressing directly on numeric values, a value LLM is trained to simulate a step of lookahead in natural language-predicting the next action, resulting state, and rationale for its value, thereby refining value estimates without any labeled data. This self-supervised procedure yields more accurate statevalue predictions, which in turn enable lightweight search algorithms to expand fewer states while maintaining strong performance. Empirically, STL-trained value models built on moderately sized (8B parameter) open-weight LLMs boost web agent success rates by 39%, achieving comparable performance with proprietary models. STL also generalizes to multi-hop QA and math puzzles. We find that STL enables small open-source models to guide efficient search, reducing inference costs by integrating explicit reasoning with value learning. In this paper, we introduce Self-Taught Lookahead (STL), a Reward and Demo Free , selfsupervised framework for interactive, multi-step reasoning tasks. Building on evidence that search quality is strongly influenced by the accuracy of the state-value estimator [10, 37] , STL improves an 39th Conference on Neural Information Processing Systems (NeurIPS 2025).