Top news
Investing in the public benefit OfficialAnthropic NewsOct 18Popular Musk Pledges SpaceX Shares to ‘Every Kid in America’ WebBloomberg TechnologyOct 19Popular Websites are closing their doors to AI agents WebThe Verge AIOct 17 This is the new ChatGPT Voice, powered by GPT Live OfficialOpenAI on YouTubeOct 18 How AI giants will reshape global philanthropy WebAxios AIOct 19 Gemini 4 Argon Really Surprised Me WebTwo Minute PapersOct 19 Google Gemini 4 agents just beat Minecraft with no human intervention WebEngadget AIOct 19 Who could it be? 🧐 OfficialGoogle on YouTubeOct 18 Anthropic gives outside AI watchdog “employee-like” access WebArs Technica AIOct 18 What are self-replicating ‘AI worms’, and why is Anthropic worried about them? WebThe Guardian AIOct 18
GlossaryModelsIntermediate

Reasoning

Also written: Reasoner, Hybrid reasoning, Hybrid reasoning model, Neurosymbolic AI, Reasoning traces, Chain of thought, Thought process, Advanced reasoning

The ability to think logically, solve problems, and make decisions by processing information and drawing conclusions.

In plain terms

Reasoning is how AI systems think through problems. It's like following steps to figure something out, similar to how humans solve puzzles or make plans.

For example, if you ask an AI to plan a trip, it needs to reason about distances, travel times, and costs to come up with a sensible itinerary. Advanced AI models are getting better at this kind of logical thinking.

The expert version

Reasoning in AI refers to the capability of a model to process information, infer relationships, and derive logical conclusions or judgments. This involves understanding context, applying rules, and performing multi-step deductions to achieve a desired outcome.

Modern large language models exhibit emergent reasoning abilities, often through techniques like chain-of-thought prompting, which encourages the model to output intermediate steps. Tradeoffs include computational cost and the potential for logical fallacies or 'hallucinations' if the reasoning process is flawed.

Why it matters now

Frontier AI labs are focusing on advanced reasoning to move beyond simple pattern matching. New models are designed to handle more complex tasks, requiring deeper analytical capabilities and the ability to show their work. This push for more sophisticated thinking aims to make AI more reliable and useful for challenging problems.

Where it shows up 25 items, newest first
How each company uses it
Anthropic Claude
Anthropic's Claude models are incorporating reasoning traces to make their decision-making process more transparent.
Google Gemini
Google Gemini is developing models like Gemini Flash with advanced multimodal reasoning and features like Deep Think for complex problem-solving.
OpenAI ChatGPT
OpenAI is enhancing its agents and coding capabilities, suggesting a focus on improving the reasoning abilities of its models.
Related terms

Problem SolvingLogicInferencePlanningChain-of-ThoughtMultimodal AI

Often confused with
Understanding
Understanding is about grasping meaning, while reasoning is about
Memory
Memory is storing information, whereas reasoning is actively using that
Pattern Recognition
Pattern recognition identifies similarities, but reasoning
Origin

The concept of reasoning in AI draws from classical logic and cognitive science, aiming to replicate human deductive and inductive processes. Early AI research focused on symbolic reasoning, while modern approaches leverage neural networks to learn reasoning patterns from data. Source ↗

Try this about 5 minutes

Ask an AI chatbot to explain a simple logical puzzle, like a riddle, and then ask it to explain its step-by-step thinking process. Observe how it breaks down the problem and arrives at the answer. You can also ask it to compare two similar items and explain the key differences, which requires analytical reasoning.

Try it: Playbook entries that use thisPlaybook →

The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.

Questions people ask

What is the difference between reasoning and just answering a question?
Answering a question can sometimes be a direct retrieval of information. Reasoning involves a deeper process of thinking, analyzing, and inferring to arrive at an answer, especially for complex or novel questions.
Can AI truly reason like humans?
Current AI models can mimic aspects of human reasoning through sophisticated pattern matching and logical deduction learned from vast datasets. However, whether this constitutes true consciousness or understanding is a subject of ongoing debate.

First seen on Subvolts Oct 29, 2021. Updated Oct 12, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.