# Fork-Think with Confidence Source: [Fork-Think with Confidence]() Collection index: [llms.txt]() Data updated: 2026-07-02T05:00:13.060Z Authors: Zena Al-Khalili, Rafi Hakim, Dietrich Klakow, Ji-Ung Lee Published: 2026-06-30T10:59:20.000Z ## Abstract Parallel thinking has enjoyed great success for boosting LLM performance on reasoning tasks without the need for any re-training. However, existing methods follow a think-first-then-decide paradigm, i.e., they first sample multiple reasoning paths, which inevitably leads to overgeneration, then prune or stop unnecessary paths to compensate. In contrast, decide-first-then-think, i.e., first identifying points that are likely to lead to desirable generations, has been underexplored so far. Following this paradigm, we propose Fork-think with confidence, that first identifies forking points using model confidence in a single seeding path, then triggers thinking, sampling multiple continuations and aggregating them for the final response. Our experiments across three models and three reasoning benchmarks show that Fork-think reduces the token consumption by up to 30% and run-time by up to 57%, while performing comparable to or better than parallel thinking. Our analysis reveals that Fork-think is able to identify forking points that are meaningful with respect to the downstream task and that sampling at later positions can lead to substantially better generations. Finally, we demonstrate how combining Fork-think with existing mechanisms such as early stopping and weighted voting can further boost the performance and perform comparably to existing state-of-the-art methods, without requiring any warm-up or offline training. Our results establish pre-determined forking as a promising research direction for efficient LLM reasoning. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Question Answering on GPQA Diamond (Accuracy, Tokens, Token Reduction)](): Question Answering; GPQA Diamond; Accuracy: 70.1. - [Mathematical Reasoning on AIME24 (Accuracy, Tokens, Token Reduction %)](): Mathematical Reasoning; AIME24; Accuracy: 86. - [Mathematical Reasoning on AIME 25 (Accuracy, Tokens, Token Reduction %)](): Mathematical Reasoning; AIME 25; Accuracy: 82.4. - [Scientific Reasoning on GPQA Diamond (Accuracy MV, Accuracy WMV, Required Tokens)](): Scientific Reasoning; GPQA Diamond; Accuracy (MV): 70.7. - [Mathematical Reasoning on AIME 24 (MV/WMV Acc, Tokens)](): Mathematical Reasoning; AIME 24; Accuracy (MV): 86. - [Mathematical Reasoning on AIME25 (Accuracy MV, Accuracy WMV, Required Tokens)](): Mathematical Reasoning; AIME25; Accuracy (MV): 82.4. ## Original sources and implementation [Original paper]() [PDF]()