# Learning a Continue-Thinking Token for Enhanced Test-Time Scaling

arXiv: https://arxiv.org/abs/2506.11274 · HTML: https://mlsanity.com/paper/2506.11274/

## Key facts

- Cited by 4 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 20 of its 33 references that are in the map.
- First posted on arXiv 12 Jun 2025.

## Placed in

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

## Builds on (most cited first, 20 of 20)

- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters](https://mlsanity.com/paper/2408.03314/) (2024)
- [Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them](https://mlsanity.com/paper/2210.09261/) (2022)
- [s1: Simple test-time scaling](https://mlsanity.com/paper/2501.19393/) (2025)
- [Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvement](https://mlsanity.com/paper/2409.12122/) (2024)
- [Tulu 3: Pushing Frontiers in Open Language Model Post-Training](https://mlsanity.com/paper/2411.15124/) (2024)
- [SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training](https://mlsanity.com/paper/2501.17161/) (2025)
- [L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning](https://mlsanity.com/paper/2503.04697/) (2025)
- [Reinforcement Learning for Reasoning in Large Language Models with One Training Example](https://mlsanity.com/paper/2504.20571/) (2025)
- [Let's Think Dot by Dot: Hidden Computation in Transformer Language Models](https://mlsanity.com/paper/2404.15758/) (2024)
- [Crossing the Reward Bridge: Expanding RL with Verifiable Rewards Across Diverse Domains](https://mlsanity.com/paper/2503.23829/) (2025)
- [From Decoding to Meta-Generation: Inference-time Algorithms for Large Language Models](https://mlsanity.com/paper/2406.16838/) (2024)
- [Scaling Test-Time Compute Without Verification or RL is Suboptimal](https://mlsanity.com/paper/2502.12118/) (2025)
- [Guiding Language Model Reasoning with Planning Tokens](https://mlsanity.com/paper/2310.05707/) (2023)
- [Do Large Language Model Benchmarks Test Reliability?](https://mlsanity.com/paper/2502.03461/) (2025)
- [m1: Unleash the Potential of Test-Time Scaling for Medical Reasoning with Large Language Models](https://mlsanity.com/paper/2504.00869/) (2025)
- ["Well, Keep Thinking": Enhancing LLM Reasoning with Adaptive Injection Decoding](https://mlsanity.com/paper/2503.10167/) (2025)

## Built on by (most cited first, 4 of 4)

- [Soft Tokens, Hard Truths](https://mlsanity.com/paper/2509.19170/) (2025)
- [What Characterizes Effective Reasoning? Revisiting Length, Review, and Structure of CoT](https://mlsanity.com/paper/2509.19284/) (2025)
- [Know Thy Reasoner: Not All Language Models Explore Alike](https://mlsanity.com/paper/2604.10827/) (2026)
- [Inject a Turn, Not a Token: Learning Where to Redirect LLM Reasoning for Self-Correction](https://mlsanity.com/paper/2610.13672/) (2026)

## Abstract (arXiv)

Test-time scaling has emerged as an effective approach for improving language model performance by utilizing additional compute at inference time. Recent studies have shown that overriding end-of-thinking tokens (e.g., replacing "</think>" with "Wait") can extend reasoning steps and improve accuracy. In this work, we explore whether a dedicated continue-thinking token can be learned to trigger extended reasoning. We augment a distilled version of DeepSeek-R1 with a single learned "<|continue-thinking|>" token, training only its embedding via reinforcement learning while keeping the model weights frozen. Our experiments show that this learned token achieves improved accuracy on standard math benchmarks compared to both the baseline model and a test-time scaling approach that uses a fixed token (e.g., "Wait") for budget forcing. In particular, we observe that in cases where the fixed-token approach enhances the base model's accuracy, our method achieves a markedly greater improvement. For example, on the GSM8K benchmark, the fixed-token approach yields a 1.3% absolute improvement in accuracy, whereas our learned-token method achieves a 4.2% improvement over the base model that does not use budget forcing.