# Base Models Know How to Reason, Thinking Models Learn When

arXiv: https://arxiv.org/abs/2510.07364 · HTML: https://mlsanity.com/paper/2510.07364/

## Key facts

- Cited by 27 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 38 of its 58 references that are in the map.
- First posted on arXiv 8 Oct 2025.

## Placed in

- LLM reasoning / [Reasoning benchmarks and evaluation](https://mlsanity.com/lr/reasoning-benchmarks-and-evaluation/)

## Builds on (most cited first, 30 of 38)

- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [Measuring Mathematical Problem Solving With the MATH Dataset](https://mlsanity.com/paper/2103.03874/) (2021)
- [MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark](https://mlsanity.com/paper/2406.01574/) (2024)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [s1: Simple test-time scaling](https://mlsanity.com/paper/2501.19393/) (2025)
- [Representation Engineering: A Top-Down Approach to AI Transparency](https://mlsanity.com/paper/2310.01405/) (2023)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?](https://mlsanity.com/paper/2504.13837/) (2025)
- [Steering Language Models With Activation Engineering](https://mlsanity.com/paper/2308.10248/) (2023)
- [Show Your Work: Scratchpads for Intermediate Computation with Language Models](https://mlsanity.com/paper/2112.00114/) (2021)
- [Scaling and evaluating sparse autoencoders](https://mlsanity.com/paper/2406.04093/) (2024)
- [Open-Reasoner-Zero: An Open Source Approach to Scaling Up Reinforcement Learning on the Base Model](https://mlsanity.com/paper/2503.24290/) (2025)
- [Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRs](https://mlsanity.com/paper/2503.01307/) (2025)
- [Reinforcement Learning for Reasoning in Large Language Models with One Training Example](https://mlsanity.com/paper/2504.20571/) (2025)
- [SciBench: Evaluating College-Level Scientific Problem-Solving Abilities of Large Language Models](https://mlsanity.com/paper/2307.10635/) (2023)
- [k-Sparse Autoencoders](https://mlsanity.com/paper/1312.5663/) (2013)
- [TheoremQA: A Theorem-driven Question Answering dataset](https://mlsanity.com/paper/2305.12524/) (2023)
- [Echo Chamber: RL Post-training Amplifies Behaviors Learned in Pretraining](https://mlsanity.com/paper/2504.07912/) (2025)
- [Thought Anchors: Which LLM Reasoning Steps Matter?](https://mlsanity.com/paper/2506.19143/) (2025)
- [Sparse Autoencoders Trained on the Same Data Learn Different Features](https://mlsanity.com/paper/2501.16615/) (2025)
- [DeepSeek-R1 Thoughtology: Let's think about LLM Reasoning](https://mlsanity.com/paper/2504.07128/) (2025)
- [NaturalReasoning: Reasoning in the Wild with 2.8M Challenging Questions](https://mlsanity.com/paper/2502.13124/) (2025)
- [Understanding Reasoning in Thinking Language Models via Steering Vectors](https://mlsanity.com/paper/2506.18167/) (2025)
- [Towards a Mechanistic Interpretation of Multi-Step Reasoning Capabilities of Language Models](https://mlsanity.com/paper/2310.14491/) (2023)
- [Inverse Scaling in Test-Time Compute](https://mlsanity.com/paper/2507.14417/) (2025)
- [I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoders](https://mlsanity.com/paper/2503.18878/) (2025)
- [Overcoming Sparsity Artifacts in Crosscoders to Interpret Chat-Tuning](https://mlsanity.com/paper/2504.02922/) (2025)
- [Steering LLM Thinking with Budget Guidance](https://mlsanity.com/paper/2506.13752/) (2025)

## Built on by (most cited first, 27 of 27)

- [Reasoning Theater: Disentangling Model Beliefs from Chain-of-Thought](https://mlsanity.com/paper/2603.05488/) (2026)
- [How do LLMs Compute Verbal Confidence](https://mlsanity.com/paper/2603.17839/) (2026)
- [Verifying Chain-of-Thought Reasoning via Its Computational Graph](https://mlsanity.com/paper/2510.09312/) (2025)
- [Measuring Chain-of-Thought Monitorability Through Faithfulness and Verbosity](https://mlsanity.com/paper/2510.27378/) (2025)
- [Reasoning Models Generate Societies of Thought](https://mlsanity.com/paper/2601.10825/) (2026)
- [Triggering Chain-of-Thought via Latent Feature Interventions in Large Language Models](https://mlsanity.com/paper/2601.08058/) (2026)
- [RISER: Orchestrating Latent Reasoning Skills for Adaptive Activation Steering](https://mlsanity.com/paper/2601.09269/) (2026)
- [Towards a Mechanistic Understanding of Large Reasoning Models: A Survey of Training, Inference, and Failures](https://mlsanity.com/paper/2601.19928/) (2026)
- [State over Tokens: Characterizing the Role of Reasoning Tokens](https://mlsanity.com/paper/2512.12777/) (2025)
- [Modeling Hierarchical Thinking in Large Reasoning Models](https://mlsanity.com/paper/2510.22437/) (2025)
- [How LLMs Detect and Correct Their Own Errors: The Role of Internal Confidence Signals](https://mlsanity.com/paper/2604.22271/) (2026)
- [Neural Procedural Memory: Empowering LLM Agents with Implicit Activation Steering](https://mlsanity.com/paper/2606.29824/) (2026)
- [Reasoning Can Be Restored by Correcting a Few Decision Tokens](https://mlsanity.com/paper/2605.16874/) (2026)
- [Transcoder Adapters for Reasoning-Model Diffing](https://mlsanity.com/paper/2602.20904/) (2026)
- [The Master Key Hypothesis: Unlocking Cross-Model Capability Transfer via Linear Subspace Alignment](https://mlsanity.com/paper/2604.06377/) (2026)
- [Position: On the Methodological Pitfalls of Evaluating Base LLMs for Reasoning](https://mlsanity.com/paper/2511.10381/) (2025)
- [Step-Tagging: Toward controlling the generation of Language Reasoning Models through step monitoring](https://mlsanity.com/paper/2512.14332/) (2025)
- [The Deepfake Detective: Interpreting Neural Forensics Through Sparse Features and Manifolds](https://mlsanity.com/paper/2512.21670/) (2025)
- [TRACES: Tagging Reasoning Steps for Adaptive Cost-Efficient Early-Stopping](https://mlsanity.com/paper/2604.21057/) (2026)
- [Latent Reward Steering: An Adaptive Inference-Time Framework that Implicitly Promotes Cognitive Behaviors in Reasoning LLMs](https://mlsanity.com/paper/2606.00726/) (2026)
- [Reasoning Fine-Tuning Induces Persistent Latent Policy States](https://mlsanity.com/paper/2607.18532/) (2026)
- [On the Generalization of Steering Vectors for Chain-of-Thought Faithfulness](https://mlsanity.com/paper/2607.29062/) (2026)
- [Math Reasoning in LLMs is Organized by Approach, Not Topic](https://mlsanity.com/paper/2609.27041/) (2026)
- [Fine Until Fine-Tuned: Repeated Solutions Make Reasoning Fragile](https://mlsanity.com/paper/2609.33559/) (2026)
- [A Polyphonic Conception of AI Understanding](https://mlsanity.com/paper/2609.36079/) (2026)
- [The Thinking Leakage in Reinforcement Learning for LLMs](https://mlsanity.com/paper/2610.14270/) (2026)
- [SFT Leaves a Reproducible Signature in Activation Geometry: Calibrating Post-Training Interpretability Against Run-to-Run Noise](https://mlsanity.com/paper/2610.17093/) (2026)

## Abstract (arXiv)

What do thinking language models learn during training that their base models lack? We first present an unsupervised method that discovers a model's reasoning behaviors by training small Sparse Autoencoders on sentence-level activations of reasoning traces, yielding interpretable reasoning taxonomies. Building on this, we introduce constructive model diffing, which aims to reconstruct the base-to-fine-tuned difference from interpretable components: reasoning mechanisms (category vectors that can induce a reasoning behavior in the base model) and reasoning heuristics (a classifier determining when a mechanism should fire). Across nine base/thinking pairs (four RL-trained, four SFT-distilled, one mixed), two independent findings agree: category vectors in the base model converge to far lower loss for taxonomies derived from purely RL-trained models, and hybrid models recover roughly 76% of the RL base-to-thinking gap but only 11% of the SFT gap. This indicates RL primarily teaches heuristics for orchestrating pre-existing base mechanisms, whereas SFT-distillation installs new ones, offering a new lens on what training paradigms teach, with implications for efficient reasoning-model development.