# Scale or Reason? A Compute-Equivalent Analysis of Reasoning Distillation

arXiv: https://arxiv.org/abs/2509.22193 · HTML: https://mlsanity.com/paper/2509.22193/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning.
- Builds on 21 of its 46 references that are in the map.
- First posted on arXiv 26 Sep 2025.

## Placed in

- LLM reasoning / [Prompting and chain-of-thought reasoning](https://mlsanity.com/lr/prompting-and-chain-of-thought-reasoning/)
- Knowledge distillation / [Knowledge distillation for LLMs](https://mlsanity.com/distill/knowledge-distillation-for-llms/)

## Builds on (most cited first, 21 of 21)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Finetuned Language Models Are Zero-Shot Learners](https://mlsanity.com/paper/2109.01652/) (2021)
- [Training Compute-Optimal Large Language Models](https://mlsanity.com/paper/2203.15556/) (2022)
- [Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters](https://mlsanity.com/paper/2408.03314/) (2024)
- [Decoupled Weight Decay Regularization](https://mlsanity.com/paper/1711.05101/) (2017)
- [gpt-oss-120b & gpt-oss-20b Model Card](https://mlsanity.com/paper/2508.10925/) (2025)
- [Instruction-Following Evaluation for Large Language Models](https://mlsanity.com/paper/2311.07911/) (2023)
- [Large Language Monkeys: Scaling Inference Compute with Repeated Sampling](https://mlsanity.com/paper/2407.21787/) (2024)
- [SciBench: Evaluating College-Level Scientific Problem-Solving Abilities of Large Language Models](https://mlsanity.com/paper/2307.10635/) (2023)
- [Llama-Nemotron: Efficient Reasoning Models](https://mlsanity.com/paper/2505.00949/) (2025)
- [Retrieval-Augmented Generation for Code Summarization via Hybrid GNN](https://mlsanity.com/paper/2006.05405/) (2020)
- [Don't Overthink it. Preferring Shorter Thinking Chains for Improved LLM Reasoning](https://mlsanity.com/paper/2505.17813/) (2025)
- [Magistral](https://mlsanity.com/paper/2506.10910/) (2025)
- [Infinity Instruct: Scaling Instruction Selection and Synthesis to Enhance Language Models](https://mlsanity.com/paper/2506.11116/) (2025)
- [Power Scheduler: A Batch Size and Token Number Agnostic Learning Rate Scheduler](https://mlsanity.com/paper/2408.13359/) (2024)
- [TeacherLM: Teaching to Fish Rather Than Giving the Fish, Language Modeling Likewise](https://mlsanity.com/paper/2310.19019/) (2023)

## Built on by (most cited first, 3 of 3)

- [KORMo: Korean Open Reasoning Model for Everyone](https://mlsanity.com/paper/2510.09426/) (2025)
- [Colon-X: Advancing Intelligent Colonoscopy toward Clinical Reasoning](https://mlsanity.com/paper/2512.03667/) (2025)
- [Building Multilingual Bridges: Data Mixing as the Pillar of Generalization for In-Language Reasoning](https://mlsanity.com/paper/2609.10445/) (2026)

## Abstract (arXiv)

Distilling reasoning traces from strong teacher models has become the standard recipe for building capable small language models. Yet reasoning traces are 5-20× longer than standard instruction fine-tuning (IFT) outputs, meaning every practitioner who chooses reasoning distillation implicitly forgoes training a larger IFT model on the same compute budget. Whether this trade-off is worthwhile remains unaddressed. We study it with a controlled experiment: a single teacher generates paired IFT and reasoning outputs for identical prompts by toggling only its reasoning mode, isolating supervision format as the sole variable. Training students at five scales (0.5B to 14B) and evaluating on 18 benchmarks, we find that at matched FLOPs, IFT lies on or near the Pareto frontier across the majority of configurations. Reasoning reaches the Pareto frontier only on open-ended tasks at 7B and above. Even there, a sequential curriculum mixing just 25-50% reasoning data with IFT captures most of the accuracy benefit at far lower compute cost.