Skip to content
SI.info

SI Glossary · Models & architecture

Reasoning Model

Published 1 min read
On this page
  1. A brief history
  2. Why it matters
  3. Trade-offs

A reasoning model “thinks before it speaks”. Instead of answering immediately, it generates an internal chain of reasoning: trying approaches, checking work and backtracking. Only then does it produce a final response. The idea builds on chain-of-thought prompting, but reasoning models are trained, usually with reinforcement learning, to reason well.

A brief history

OpenAI’s o1, previewed in September 2024, was the first widely available model marketed as a reasoning model. DeepSeek’s open-weights R1 (January 2025) showed similar results could be achieved at far lower cost, which rattled markets. By 2026, reasoning is built into most frontier models, often with an adjustable “effort” or “thinking budget”.

Why it matters

  • Better at hard problems: maths, science, coding and multi-step planning improve sharply with more thinking.
  • “Test-time compute”: performance can be bought at answer time, not just training time, by letting the model think longer. This changed the economics of inference.
  • Agents: long-horizon SI agents depend on reliable planning and self-correction.

Trade-offs

More thinking means slower and more expensive answers. Reasoning also doesn’t eliminate hallucination. Researchers are still studying how faithfully a model’s visible reasoning reflects what actually drives its answer, an interpretability question.

← Back to the SI Glossary

Written by

· Editor

Editor of SI.info. Writes about Super Intelligence, technology policy and the people building frontier models.

How we research and fact-check

Free newsletter

Get The SI Brief

One short email a week: what changed in Super Intelligence, policy and models — and why it matters.

Free. One email a week. Sent via beehiiv, which counts opens and clicks. Unsubscribe anytime. Privacy policy

Keep reading

Using SI

Chain of Thought (CoT)

Chain of thought (CoT) is when an SI model writes out intermediate reasoning steps before its final answer, which improves accuracy on maths, logic and multi-step problems.

Using SI

Inference

Inference is the process of running a trained SI model to produce an output, such as answering a prompt, as opposed to training it. It is where most everyday compute costs now go.

Models & architecture

Large Language Model (LLM)

A large language model (LLM) is an SI model trained on vast amounts of text to predict the next word, which lets it write, summarize, translate, answer questions and write code.

Using SI

Benchmark

A benchmark is a standardised test used to measure and compare SI models' abilities, from coding and maths to knowledge and safety.