---
title: "What is a reasoning model in AI? Reasoning vs LLMs"
description: "A reasoning model is an LLM trained to generate a chain of thought before answering, trading latency and tokens for accuracy on math, code, and planning tasks."
type: "glossary-term"
canonical_url: "https://theaiengineering.com/glossary/reasoning-model"
markdown_url: "https://theaiengineering.com/glossary/reasoning-model.md"
topic: "LLM models and benchmarks"
author: "The AI Engineering Editors"
publisher: "The AI Engineering"
updated: "2026-11-06"
keywords: ["Reasoning model","large reasoning model","LRM","thinking model"]
citation: "The AI Engineering Editors. \"What is a reasoning model in AI? Reasoning vs LLMs.\" The AI Engineering, updated Nov 06, 2026. https://theaiengineering.com/glossary/reasoning-model"
---

# What is a reasoning model in AI? Reasoning vs LLMs

Also called large reasoning model, LRM, thinking model.

> **Definition of Reasoning model:** A reasoning model is a large language model trained, usually with reinforcement learning, to generate intermediate reasoning tokens before its final answer, spending more inference compute to solve multi-step problems.

A reasoning model is a [large language model](https://theaiengineering.com/glossary/large-language-model) that works through a problem in an extended [chain of thought](https://theaiengineering.com/glossary/chain-of-thought) before it commits to an answer. OpenAI's o1, announced on September 12, 2024, brought the approach to a commercial API. DeepSeek-R1 followed in January 2025 with open weights, and Anthropic's Claude 3.7 Sonnet, released in February 2025, was a hybrid model that could answer directly or generate extended reasoning first. Current OpenAI and Anthropic APIs both expose controls for how much a model reasons.

## How reasoning models work

Two ideas combine: training the model to reason, and letting it spend more compute on hard problems at inference time.

- **Training.** Reinforcement learning rewards the model for reaching correct answers on problems that can be checked automatically, an approach often called reinforcement learning with verifiable rewards (see [RLHF](https://theaiengineering.com/glossary/rlhf)). The [DeepSeek-R1 paper](https://arxiv.org/abs/2501.12948) reports that rule-based rewards for answer accuracy and output format, applied at scale, produced behaviors such as self-verification and reflection without human-written reasoning examples.
- **Inference.** The model generates more [tokens](https://theaiengineering.com/glossary/tokens) of reasoning on harder problems. OpenAI reported that o1's performance improved both with more reinforcement learning during training and with more compute spent on reasoning at test time.

APIs expose a control for this trade-off, such as OpenAI's `reasoning.effort` parameter or Anthropic's effort setting. According to [OpenAI's reasoning guide](https://developers.openai.com/api/docs/guides/reasoning), reasoning tokens are billed as output tokens and occupy context window space even though the API does not return them.

## Reasoning models vs standard LLMs

| | Standard LLM | Reasoning model |
|---|---|---|
| Output | Answer directly | Reasoning tokens, then the answer |
| [Time to first token](https://theaiengineering.com/glossary/time-to-first-token) of the answer | Low | Higher, and grows with reasoning length |
| Cost per request | Lower | Higher, since reasoning is billed as output |
| Strongest on | Chat, rewriting, extraction, lookups | Math, code, multi-step planning, agents |

Reasoning models are not the right choice for every request. For simple lookups, classification, or latency-sensitive chat, a standard model or a low effort setting is usually the better fit. Reasoning ability can also be [distilled](https://theaiengineering.com/glossary/knowledge-distillation) into smaller models: DeepSeek fine-tuned Qwen and Llama models from 1.5B to 70B parameters on DeepSeek-R1 outputs. For more depth, see the [reasoning models explainer](https://theaiengineering.com/articles/reasoning-models) and the [LLM benchmarks explainer](https://theaiengineering.com/articles/llm-benchmarks).

## Related terms

- [RLHF](https://theaiengineering.com/glossary/rlhf): RLHF (reinforcement learning from human feedback) is a fine-tuning method that trains a reward model on human rankings of model outputs, then optimizes the language model with reinforcement learning to score highly.
- [Knowledge distillation](https://theaiengineering.com/glossary/knowledge-distillation): Knowledge distillation is a model compression technique in which a smaller student model is trained to reproduce the outputs or output probabilities of a larger, more capable teacher model.
- [Tokens](https://theaiengineering.com/glossary/tokens): Tokens are the units of text a language model processes, typically whole words, word fragments, punctuation, or bytes, each mapped to an integer ID from the model's fixed vocabulary.
- [Time to first token](https://theaiengineering.com/glossary/time-to-first-token): Time to first token (TTFT) is the elapsed time from submitting a request to an LLM until the first token of the response arrives, covering request queuing, network transit, and prompt prefill.
- [Large language model (LLM)](https://theaiengineering.com/glossary/large-language-model): A large language model (LLM) is a neural network, usually a transformer with billions of parameters, trained on web-scale text to predict the next token and then adapted to follow instructions.
- [Test-time compute](https://theaiengineering.com/glossary/test-time-compute): Test-time compute is the computation a language model spends while answering a query, through longer reasoning, multiple samples, or search, as opposed to the train-time compute spent building the model.
- [Chain-of-thought prompting](https://theaiengineering.com/glossary/chain-of-thought): Chain-of-thought prompting is a technique that has a language model write intermediate reasoning steps before its final answer, which improves large models' accuracy on arithmetic, commonsense, and symbolic reasoning tasks.

## Guides that cover reasoning model

- [Reasoning models: how they work and when to use them](https://theaiengineering.com/articles/reasoning-models): Reasoning models explained: how test-time compute and reinforcement learning produce chain-of-thought reasoning, when to use them, and their cost and latency.
- [LLM benchmarks explained: what they measure and how to read them](https://theaiengineering.com/articles/llm-benchmarks): LLM benchmarks explained: what MMLU-Pro, GPQA, SWE-bench, LiveCodeBench, ARC-AGI, and LMArena measure, why scores saturate, and how to read a leaderboard.

Part of the [LLM models and benchmarks](https://theaiengineering.com/topics/models) topic hub.

## Cite this definition

The AI Engineering Editors. "What is a reasoning model in AI? Reasoning vs LLMs." The AI Engineering, updated Nov 06, 2026. https://theaiengineering.com/glossary/reasoning-model

- Title: What is a reasoning model in AI? Reasoning vs LLMs
- Author: The AI Engineering Editors
- Publisher: The AI Engineering
- Updated: 2026-11-06
- URL: https://theaiengineering.com/glossary/reasoning-model
- Markdown: https://theaiengineering.com/glossary/reasoning-model.md
