# LYNX: Learning Dynamic Exits for Confidence-Controlled Reasoning

arXiv: https://arxiv.org/abs/2512.05325 · HTML: https://mlsanity.com/paper/2512.05325/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning.
- Builds on 19 of its 28 references that are in the map.
- First posted on arXiv 5 Dec 2025.

## Placed in

- LLM reasoning / [Test-time compute and efficient thinking](https://mlsanity.com/lr/test-time-compute-and-efficient-thinking/)
- Efficient LLM inference and serving / [Efficient reasoning and agent inference](https://mlsanity.com/inference/efficient-reasoning-and-agent-inference/)

## Builds on (most cited first, 19 of 19)

- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Measuring Mathematical Problem Solving With the MATH Dataset](https://mlsanity.com/paper/2103.03874/) (2021)
- [Let's Verify Step by Step](https://mlsanity.com/paper/2305.20050/) (2023)
- [Understanding intermediate layers using linear classifier probes](https://mlsanity.com/paper/1610.01644/) (2016)
- [A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification](https://mlsanity.com/paper/2107.07511/) (2021)
- [Discovering Latent Knowledge in Language Models Without Supervision](https://mlsanity.com/paper/2212.03827/) (2022)
- [Designing and Interpreting Probes with Control Tasks](https://mlsanity.com/paper/1909.03368/) (2019)
- [Dynamic Early Exit in Reasoning Models](https://mlsanity.com/paper/2504.15895/) (2025)
- [Conformal Language Modeling](https://mlsanity.com/paper/2306.10193/) (2023)
- [Reasoning Models Know When They're Right: Probing Hidden States for Self-Verification](https://mlsanity.com/paper/2504.05419/) (2025)
- [Between Underthinking and Overthinking: An Empirical Study of Reasoning Length and correctness in LLMs](https://mlsanity.com/paper/2505.00127/) (2025)
- [Wait, We Don't Need to "Wait"! Removing Thinking Tokens Improves Reasoning Efficiency](https://mlsanity.com/paper/2506.08343/) (2025)
- [How Does BERT Answer Questions? A Layer-Wise Analysis of Transformer Representations](https://mlsanity.com/paper/1909.04925/) (2019)
- [Think or Not? Exploring Thinking Efficiency in Large Reasoning Models via an Information-Theoretic Lens](https://mlsanity.com/paper/2505.18237/) (2025)
- [FlashThink: An Early Exit Method For Efficient Reasoning](https://mlsanity.com/paper/2505.13949/) (2025)
- [The Trilemma of Truth in Large Language Models](https://mlsanity.com/paper/2506.23921/) (2025)
- [The Zero-Step Thinking: An Empirical Study of Mode Selection as Harder Early Exit in Reasoning Models](https://mlsanity.com/paper/2510.19176/) (2025)

## Built on by (most cited first, 6 of 6)

- [Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models](https://mlsanity.com/paper/2605.17672/) (2026)
- [Mitigating Overthinking in Large Reasoning Language Models via Reasoning Path Deviation Monitoring](https://mlsanity.com/paper/2603.14251/) (2026)
- [Rethinking RL for LLM Reasoning: It's Sparse Policy Selection, Not Capability Learning](https://mlsanity.com/paper/2605.06241/) (2026)
- [Know When To Fold 'Em: Token-Efficient LLM Synthetic Data Generation via Multi-Stage In-Flight Rejection](https://mlsanity.com/paper/2605.14062/) (2026)
- [BLADE: Boundary-Expanded and Layer-Adaptive Dynamic Exit for Efficient LLM Reasoning](https://mlsanity.com/paper/2607.28966/) (2026)
- [Settle: Learning When to Stop Reasoning](https://mlsanity.com/paper/2609.38997/) (2026)

## Abstract (arXiv)

Large reasoning models achieve strong performance on complex tasks by generating extended chains of thought, but they often "overthink": continuing to reason long after they have enough information to answer correctly. This wastes inference-time compute and can hurt accuracy. Existing attempts to stop early either manipulate decoding with extra sampling and heuristics, rely on auxiliary verifier models, or operate only as post-hoc analysis pipelines without formal guarantees. We introduce LYNX, an online early-exit mechanism that turns a model's own hidden-state awareness into confidence-controlled stopping decisions. LYNX attaches exit decisions to naturally occurring reasoning cues (e.g., "hmm", "wait") during generation, trains a lightweight probe on hidden states at those cue tokens using supervision from forced exits, and wraps the resulting scores in split conformal prediction to obtain distribution-free control over premature exits. Crucially, we train and calibrate this probe once on a generic mathematical corpus and reuse it unchanged across benchmarks, decoding temperatures, and even non-mathematical tasks. Across three model families spanning 1.5B to 32B parameters, a single mathematically trained probe per base model yields strong accuracy--efficiency tradeoffs. On GSM8K, LYNX matches or improves baseline accuracy while reducing tokens by 40--65%; on MATH-500 it improves accuracy by up to 12 points with roughly 35--60% fewer tokens; on AIME 2024 it recovers baseline accuracy with more than 50% token savings; and on CommonsenseQA, a non-math benchmark, it transfers zero-shot with modest accuracy gains and up to 70% fewer tokens. Compared to state-of-the-art early-exit methods, LYNX offers competitive or superior Pareto frontiers while remaining fully online, requiring no proxy models at inference, and providing explicit, user-tunable confidence guarantees.