Reasoning model: A reasoning model is a large language model trained to work through a problem step by step before answering, spending extra computation (often called thinking) to improve accuracy on math, code, logic, and multi-step tasks.
A standard chat model starts writing its answer immediately. A reasoning model first generates a long internal chain of intermediate steps, checks its own work, tries alternatives, and only then writes the reply you see. Some products show a summary of that thinking; others hide it. OpenAI's o-series and GPT-5 thinking modes, Anthropic's Claude with extended thinking, Google's Gemini thinking models, and DeepSeek's R1 all work this way.
The extra thinking costs time and money. A reasoning model can take seconds to minutes to answer and uses more tokens, so it is slower and pricier per question. That trade is worth it for a tricky spreadsheet formula, a contract clause with nested conditions, a debugging session, or a plan with many dependencies. It is wasted on a quick email rewrite.
Reasoning models still make mistakes, and they can make them confidently. The visible reasoning is a useful window into how the model approached the task, but researchers have shown that it does not always reflect the real reason the model reached its answer. Treat the steps as a draft to check, not as proof.
Example at work
An operations manager pastes a 40-line shipping rate table and asks which carrier is cheapest for a 1,200-pound pallet to Denver under three accessorial scenarios. The reasoning model works through each scenario before answering, and the manager verifies the winning lane against the carrier's rate sheet before booking.
Why it matters
Most chat apps now let you choose between a fast mode and a thinking mode. Picking the right one saves you time on simple tasks and saves you errors on hard ones. When accuracy matters, switch to the reasoning mode and still verify the result.
Related terms
- Chain-of-thought promptingChain-of-thought prompting asks an AI model to work through a problem step by step before giving its final answer. Showing its reasoning tends to improve accuracy on math, logic, and multi-step tasks and makes mistakes easier to spot.
- Large language model (LLM)A large language model (LLM) is a neural network trained on enormous amounts of text to predict the next word, then tuned to follow instructions and hold a conversation. LLMs power ChatGPT, Claude, Gemini, and Microsoft Copilot.
- Reinforcement learningReinforcement learning is a way of training AI by trial and error: the system takes actions, receives rewards or penalties based on the outcomes, and adjusts its behavior to earn more reward over time.
- InferenceInference is the act of running a trained AI model to produce an output, such as answering a prompt or classifying an image. Training happens once; inference happens every time anyone uses the model, and it is what you pay for as a user.
- LatencyLatency is the delay between sending a request to an AI system and receiving its response. For language models it is usually measured as time to the first word and then words per second as the answer streams in.