o1 is OpenAI's first model in the "reasoning" series. Unlike standard LLMs that respond immediately, o1 spends time "thinking" — generating an internal chain-of-thought before producing its final answer. This deliberate reasoning process dramatically boosts performance on hard science, mathematics, and competitive programming problems.
Key capabilities
- Deliberate chain-of-thought — internal reasoning tokens (not shown to users) before output
- 200K context window — handles very long documents and codebases
- PhD-level science benchmarks — 78% on GPQA Diamond (comparable to human experts)
- Vision understanding — can reason over images and diagrams
- Complex multi-step problem solving — legal analysis, financial modelling, advanced research
What it's best for
Tasks where correctness matters more than speed: competitive mathematics, algorithm design, complex code generation, scientific reasoning, and long-horizon planning. For everyday tasks or cost-sensitive workloads, GPT-4o or GPT-4o mini are better choices.