aiqbench analysis
This paper is squarely about AI: it surveys and audits early-stopping methods for large reasoning models, the techniques that cut unnecessary chain-of-thought computation. Its most useful contribution for practitioners is the finding that efficiency gains can hide reliability problems, including correctness dropping at later exit points and continued reasoning after forced termination. Readers deploying reasoning models should treat early stopping as a cost-accuracy trade-off that needs auditing rather than a free optimization.
Early Stopping in Large Reasoning Models: A Survey and Empirical Audit of Efficiency and Reliability
arXiv:2611.00166v1 Announce Type: new Abstract: Large reasoning models can improve accuracy by generating longer reasoning traces, but some of this computation may be unnecessary. Early-stopping methods aim to stop once enough reasoning is available to answer. Evaluating these methods requires accounting for the cost of deciding when to stop and the reliability of the resulting answer. We evaluate different methods across five reasoning models on mathematical and scientific reasoning, code generation, and factuality tasks by using a shared rollout replay protocol. We introduce Target-Integrated Token Efficiency (TITE) to measure efficiency relative to the oracle across accuracy targets. We also examine correctness at later exit positions, transfer across models and tasks, post-stop reasoning continuation, and retained justification quality. In our experiments, correctness can decrease at later positions beyond what the fitted decoding-noise model predicts. Reliability checks also reveal continued reasoning after forced termination and weaker supporting reasoning in some correct early-stopping outputs. The implementation can be found in https://github.com/hung20gg/audit-early-stopping.
Abstract:Large reasoning models can improve accuracy by generating longer reasoning traces, but some of this computation may be unnecessary. Early-stopping methods aim to stop once enough reasoning is available to answer. Evaluating these methods requires accounting for the cost of deciding when to stop and the reliability of the resulting answer. We evaluate different methods across five reasoning models on mathematical and scientific reasoning, code generation, and factuality tasks by using a shared rollout replay protocol. We introduce Target-Integrated Token Efficiency (TITE) to measure efficiency relative to the oracle across accuracy targets. We also examine correctness at later exit positions, transfer across models and tasks, post-stop reasoning continuation, and retained justification quality. In our experiments, correctness can decrease at later positions beyond what the fitted decoding-noise model predicts. Reliability checks also reveal continued reasoning after forced termination and weaker supporting reasoning in some correct early-stopping outputs. The implementation can be found in this https URL.
Artificial Intelligence (cs.AI)
arXiv:2611.00166 [cs.AI]
(or arXiv:2611.00166v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2611.00166 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 249 words
Related coverage
Using Explainability as a Training-Time Reliability Signal for Efficient ECG Classification
arXiv · Jun 12, 2026
EMS: Multi-Agent Voting via Efficient Majority-then-Stopping
arXiv · Jun 16, 2026
Dual-domain fused LSTM modeling for efficient time-dependent reliability analysis
arXiv · Jul 23, 2026
ToolMenuBench: Benchmarking Tool-Menu Filtering Strategies for Reliable and Efficient LLM Agents
arXiv · Jun 16, 2026