Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

Abstract summary

Demonstrates that prompting LLMs to articulate intermediate reasoning steps before producing a final answer ('chain-of-thought prompting') dramatically improves accuracy on math, logic, and multi-step problem-solving benchmarks. The improvement scales with model size and emerges only at sufficient scale. Establishes step-by-step reasoning as a critical prompting technique and a foundation for later 'reasoning model' designs.

Our summary in our own words — see the canonical source links below for the original abstract.

Why we cite this paper

Chain-of-Thought is the technique that 'reasoning models' (OpenAI o1/o3, Anthropic Claude with extended thinking, DeepSeek R1) train into the model rather than relying on prompting alone. The category emerged directly from this paper's framing. Our /trend/agentic-ai-frameworks-2026 tracks the engineering acceleration in this category.

Key findings

• 1Chain-of-thought prompting dramatically improves LLM performance on multi-step reasoning tasks.
• 2The capability emerges only at sufficient model scale (~100B parameters).
• 3Step-by-step reasoning can be elicited via few-shot prompting without model retraining.
• 4Foundation for modern reasoning models that train extended chain-of-thought as a native capability.

Canonical sources

Related glossary terms

Frequently Asked Questions

Other research papers