Reasoning Models
definition
Reasoning models like OpenAI's o1/o3 and Anthropic's Claude with extended thinking represent a paradigm shift where models spend more computational effort at inference time to solve harder problems, trading latency and cost for significantly improved accuracy on complex tasks. Unlike standard LLMs that generate output in a single forward pass, reasoning models use chain-of-thought internally — sometimes generating thousands of "thinking tokens" before producing a response — which makes them dramatically better at multi-step logic, math, and code generation.
Reasoning models like OpenAI's o1/o3 and Anthropic's Claude with extended thinking represent a paradigm shift where models spend more computational effort at inference time to solve harder problems, trading latency and cost for significantly improved accuracy on complex tasks. Unlike standard LLMs that generate output in a single forward pass, reasoning models use chain-of-thought internally — sometimes generating thousands of "thinking tokens" before producing a response — which makes them dramatically better at multi-step logic, math, and code generation. For agentic systems, the key architectural decision is when to route tasks to a reasoning model versus a standard model: simple tool calls and classification don't benefit from extra thinking, but complex planning, debugging, and architectural decisions do. Understanding the test-time compute trade-off is essential because it directly impacts your cost model — a reasoning model may cost 10-50x more per query but solve problems that cheaper models fail at entirely. This concept connects to model selection for choosing when to use reasoning models, token economics for understanding the cost implications, and chain of thought for the prompting technique that reasoning models internalize.