AI glossary · Models
Reasoning model
In one sentence
A reasoning model is a language model trained to produce a chain of intermediate steps before giving its final answer. It does better on logic, maths and planning problems, at the cost of slower replies and more tokens used.
Also known as: large reasoning model · LRM · thinking model · reasoning LLM
What is a reasoning model?
A standard LLM starts writing its answer as soon as it gets the question. A reasoning model takes a moment first: it produces an internal draft where it breaks the problem down, tries different routes, spots mistakes and corrects them. Only then does it write the answer.
Put it to work
AI agents that look after your customers
From €49/mo + VAT
Voice, WhatsApp and email with the phone line included: agent, numbers and CRM in one fee.
See plans and pricingThe technique has a name, chain of thought, and it existed before as a prompting trick: asking the model to “think step by step”. What’s new is that reasoning models are trained specifically to do this well and on their own. The first to gain wide attention were OpenAI’s o1 series in 2024 and DeepSeek-R1 in early 2025; several providers now offer equivalent models or modes.
Related to this term
UTIO agents
- Available 24/7 in 26 languages
- Set it up yourself for €0: built from your website
- 14-day free trial, no card needed
From €49/mo + VAT
See plans and pricingHow it works
- Receive the question like any other model.
- Reason: generate a sequence, sometimes a long one, of intermediate steps. Depending on the provider, that reasoning is shown, summarised or hidden.
- Respond with the conclusion, building on those steps.
Training usually involves reinforcement learning on problems with checkable answers, such as maths or code, so the model learns which ways of reasoning lead to correct results. Many let you set the “effort”: more steps for hard problems, fewer for easy ones.
Thinking longer has a double price: more tokens generated and a longer wait.
Example
A building contractor in Cardiff prices jobs from measurements, material costs and the schedules of several trades. A standard LLM can draft the quote, but it sometimes gets quantities wrong or muddles the order of jobs that depend on each other.
A reasoning model works through the figures step by step and proposes a sensible schedule: plumbing first, then tiling, painting last. It takes longer to reply, but for a quote that’s reviewed at leisure, that extra minute doesn’t matter. The site manager checks it before it goes out.
How it affects your business
Not every task needs a reasoning model. Telling a customer whether you’re open on Saturday doesn’t call for ten steps of thought. Reviewing a contract or planning delivery routes might.
Latency is the deciding factor. With UTIO’s voice agent, the conversation happens in real time, and what your caller notices is how long it takes to reply. So for an AI agent that talks on the phone, speed matters as much as raw capability. Heavy tasks fit better outside the call.
Frequently asked questions
Does a reasoning model think like a person?
No. 'Reasoning' means it generates intermediate steps before answering, which improves its results on many problems. But it's still a model that predicts text: it can chain together steps that look logical and still reach the wrong conclusion.
Why is a reasoning model more expensive?
Because those intermediate steps are tokens the model generates, and providers usually charge for them even when they aren't shown. A short answer can sit on top of thousands of reasoning tokens.
Should a voice agent use a reasoning model?
For the conversation itself, usually not: callers notice the pauses. It fits better with tasks that aren't time-critical, such as analysing the day's calls or preparing a report, while a fast model handles the conversation.
Related terms
- Chain of thought
Chain of thought is a technique where an AI model writes out the intermediate steps of its reasoning before giving the final answer. It improves results on calculations, chained rules and multi-step problems.
- LLM
An LLM (large language model) is an AI model trained on vast amounts of text to predict the next word. That single skill lets it draft, summarise, translate, classify and hold a conversation.
- Latency
Latency is the time between making a request and starting to receive the response. In a voice agent, it's the silence from the moment the caller stops talking to the moment they hear the agent's first word, and it decides whether the call sounds natural.
- Inference
Inference is when an already-trained AI model is put to use: it receives a new input, such as a question, an audio clip or an image, and computes a response from what it learned, without changing its parameters.
- Benchmark
An AI benchmark is a standardised test, with a set of questions or tasks and a fixed scoring method, used to measure and compare how different models perform at a specific skill such as reasoning, coding or answering questions.
Updated on