reasoning model

A reasoning model, also called a large reasoning model (LRM), is a generative language or multimodal model designed and tuned to handle complex, multi-step reasoning tasks. It does that by allocating more inference resources or explicitly incorporating internal thinking phases before producing a final answer. In current practice, that reasoning is increasingly a tunable setting rather than only a fixed property of a separate model class.

The major vendors expose a reasoning effort or thinking level control on their general-purpose frontier models, and they meter the internal deliberation as reasoning tokens that count against the output budget.

Reasoning capabilities are built on top of standard LLMs through training and inference techniques, such as:

• Reinforcement learning on reasoning traces
• Explicit reasoning tokens for test-time thinking
• Search or self-consistency to improve correctness

Many reasoning models also call external tools, such as code execution, to check and extend their reasoning.

These models represent a broader shift toward models that not only generate responses, but also plan, decompose, verify, and refine reasoning internally through long chain-of-thought traces or latent internal reasoning processes.

Related Resources

Course

Accessing Multiple AI Models With the OpenRouter API

Access models from popular AI providers in Python through OpenRouter's unified API with smart routing, fallbacks, and cost controls.

intermediate ai api