Chain-of-Thought (CoT)
Prompting technique where the model is asked to reason step-by-step before answering.
Chain-of-Thought is a prompting pattern (and now a training pattern) where the model produces intermediate reasoning steps before its final answer. It significantly improves performance on math, logic, planning and multi-hop reasoning tasks.
The 2025–26 generation of models (o-series, Claude 4, Gemini 2.5 thinking modes) has CoT built into the model itself — reasoning happens automatically without prompt engineering. This dramatically raises quality on hard tasks at the cost of extra tokens and latency.
For enterprise use, CoT is best used selectively. Simple tasks don't need it; hard reasoning, multi-step planning and mathematical work do.
Key points
- Model reasons step-by-step before final answer
- Big quality lift on math, logic, planning tasks
- Modern frontier models have CoT built-in ('thinking' modes)
- Uses more tokens and adds latency — apply selectively
Common use cases
Frequently asked
Related terms
The practice of designing model inputs to produce reliable, task-specific outputs.
A neural network trained on massive text corpora to understand and generate human-like language.
An LLM-driven system that plans, calls tools, and executes multi-step tasks autonomously.
Applying Chain-of-Thought (CoT) in your business?
We help enterprises design, ship, and govern AI systems end-to-end.