Build a Reasoning Model (From Scratch)

Last updated: April 4, 2026

Introduction 

Large Language Models or LLMs, are advanced AI models trained on vast collections of text to understand and generate human-like language. They can summarize information, answer questions, write code, and adapt to diverse contexts by predicting the most relevant or coherent text output. 

A reasoning model builds upon an LLM by improving its ability to simulate structured thought - breaking down complex tasks into smaller, logical steps, verifying conclusions, and iterating to reach correct or optimized solutions. These models aim to make AI not only fluent but genuinely *thoughtful* in its approach to problem-solving.

LLM reasoning systems possess remarkable capabilities for solving complex, multi-step problems that demand logical progression and nuanced decision-making. In *Build a Reasoning Model (From Scratch)*, you’ll explore how to construct a functional reasoning model starting from the foundations upward. The process begins with a pre-trained LLM as your base, followed by the integration of specialized reasoning enhancements designed to strengthen its analytical depth and problem-solving accuracy.

Contents 

1. Understanding Reasoning Models

2. Generating Text with a Pre-Trained LLM

3. Evaluating Reasoning Models

4. Improving Reasoning with Inference-time scaling

5. Inference-time scaling via Self-refinement

6. Training Reasoning Models with Reinforcement Learning

7. Improving GRPO for Reinforcement Learning

8. Distilling Reasoning Models for Efficient Reasoning

Textbook

• This course is based on the book by Sebastian Raschka, Build a Reasoning Model (From Scratch). Published April 2026 https://www.manning.com/books/build-a-reasoning-model-from-scratch 

  • Github repo for the code: https://github.com/rasbt/reasoning-from-scratch 

  • The code can also be downloaded from: https://www.manning.com/books/build-a-reasoning-model-from-scratch (look for the 'source code' link in the left-hand column.