Reasoning model

As explained in Build a Reasoning Model (From Scratch)

A language model optimized to deliberate over complex problems, often producing intermediate steps or using tools to improve answer quality.

A reasoning model is a language model improved through training or prompting to generate intermediate problem-solving steps before its final answer. These steps may be visible, hidden in special tags, or omitted from the user-facing response. The practical goal is to improve performance on complex coding, mathematics, and logic tasks rather than to claim human-like thought. section 1.1

Use reasoning models as an engineering component when the task justifies their additional computation:

• Apply chain-of-thought or other inference techniques when a task benefits from explicit intermediate steps, decomposition, planning, or tool selection. section 1.1 section 1.6
• Prefer conventional models for simpler summarization, translation, or knowledge-based questions when longer outputs, higher cost, latency, or overthinking are poor trade-offs. section 1.6
• Distinguish deliberate reasoning behavior from implicit pattern matching: a conventional model can appear logical without explicitly tracking contradictions or applying rules. section 1.5

Math verification workflow

A model generates a free-form solution and final answer, a verifier extracts and normalizes that answer, and an external calculator or code interpreter compares it with the reference solution. The resulting correctness signal can evaluate performance and later support reinforcement-learning training. section 3.1

Figure 1.9 A high-level road map of what we’ll build in this book. We’ll start with a conventional LLM, add evaluation methods so we can measure its progress, and then explore two broad families of reasoning improvements: inference techniques and training techniques.

Use the roadmap to trace the build from a pretrained base LLM through evaluation, inference-time improvements, and training-time methods that produce a dedicated reasoning model. section 1.7

How Build a Reasoning Model (From Scratch) covers it

How deep it goes

Defines visible and hidden intermediate steps, contrasts pattern matching with explicit logic, and outlines inference and training methods without implementing them.

Examples

Uses contradictory bird premises, chain-of-thought diagrams, token-by-token generation, roadmap figures, and math-verifier workflows.

What it assumes

Starts from a pretrained conventional LLM and adds evaluation before applying inference-time or training-time reasoning improvements.

What you can do after

Evaluate/understand when intermediate reasoning improves accuracy and how verification can measure or train reasoning behavior.

How it connects

Connects chain-of-thought, inference-time methods, evaluation, verifiers, reinforcement learning, distillation, and pretrained language models.

liveBookmanning.com

Read Build a Reasoning Model (From Scratch) on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Reinforcement Learning from Human FeedbackA reasoning language model is a language model optimized to solve problems through explicit or implicit multi-step reasoning before producing a final answer.
• 

  liveBookmanning.com
  AI Agents in Action, Second EditionA reasoning model is a language model that explicitly allocates computation to multistep deliberation before producing its final output.
• 

  liveBookmanning.com
  Introduction to Generative AI, Second EditionA reasoning model is a language model that generates intermediate reasoning steps before producing a final answer, especially for difficult problems.

Compare how 4 books explain reasoning model →

Related concepts

• large language modelkind of · A large language model is a machine-learning model, typically neural, trained on extensive text data to process and generate language.
• language modelkind of · A language model is a computational model that represents patterns in linguistic tokens to estimate sequence probabilities and predict or generate language from context.
• multi-step reasoninguses · Multi-step reasoning derives a conclusion, solution, or decision through multiple connected intermediate inferential or problem-solving steps.
• chain-of-thought promptingcontrasts with · Chain-of-thought prompting is a technique that elicits or guides a model to produce or follow intermediate reasoning steps before a final answer.

This explainer was generated by Manning's knowledge graph from the text of Build a Reasoning Model (From Scratch). The section links point to the parts of the book it draws on.