Correct, Concise and Complete: Multi-stage Training For Adaptive Reasoning

Nathanaël Carraz Rakotonirina, Ren Pang, Neha Anna John, Michael Bohlke-Schneider, Momchil Hardalov


Abstract
The reasoning capabilities of large language models (LLMs) have improved substantially through increased test-time computation, typically in the form of intermediate tokens known as chain-of-thought (CoT). However, CoT often becomes unnecessarily long, increasing computation costs without improving accuracy and sometimes even degrading performance, a phenomenon known as “overthinking”. We propose a multi-stage efficient reasoning method that combines supervised fine-tuning—via rejection sampling or reasoning trace reformatting—with reinforcement learning using an adaptive length penalty. We introduce a lightweight reward function that penalizes tokens generated after the first correct answer, encouraging the model to perform self-verification only when beneficial. We conduct a holistic evaluation across seven diverse reasoning tasks, analyzing the accuracy–response length trade-off. Our approach reduces response length by an average of 28% for 8B models and 40% for 32B models, while incurring only minor performance drops of 1.6 and 2.5 points, respectively. Despite its conceptual simplicity, it achieves a better trade-off than more complex state-of-the-art efficient reasoning methods, scoring 76.6 on the area under the Overthinking-Adjusted Accuracy curve (AUCOAA)—5 points above the base model and 2.5 points above the second-best approach.
Anthology ID:
2026.findings-acl.622
Volume:
Findings of the Association for Computational Linguistics: ACL 2026
Month:
July
Year:
2026
Address:
San Diego, California, United States
Editors:
Maria Liakata, Viviane P. Moreira, Jiajun Zhang, David Jurgens
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
12798–12810
Language:
URL:
https://aclanthology.org/2026.findings-acl.622/
DOI:
Bibkey:
Cite (ACL):
Nathanaël Carraz Rakotonirina, Ren Pang, Neha Anna John, Michael Bohlke-Schneider, and Momchil Hardalov. 2026. Correct, Concise and Complete: Multi-stage Training For Adaptive Reasoning. In Findings of the Association for Computational Linguistics: ACL 2026, pages 12798–12810, San Diego, California, United States. Association for Computational Linguistics.
Cite (Informal):
Correct, Concise and Complete: Multi-stage Training For Adaptive Reasoning (Rakotonirina et al., Findings 2026)
Copy Citation:
PDF:
https://aclanthology.org/2026.findings-acl.622.pdf
Checklist:
 2026.findings-acl.622.checklist.pdf