o1
Reasoning model Generally availableOpenAI advanced reasoning model trained with reinforcement learning to generate deliberative chain-of-thought solutions for complex mathematics, coding, and scientific research.
Overview
OpenAI o1 is a specialized reasoning model designed to solve complex multi-step problems through deliberative chain-of-thought processing. Released in preview in September 2024 and generally available in December 2024, o1 represents an architectural shift toward inference-time compute scaling, spending additional time pondering before emitting final responses.
The model features an expanded 200,000-token context window and supports up to 100,000 output tokens per completion, which accommodates both internal reasoning tokens and final output generation. Unlike standard autoregressive models, o1 breaks down difficult queries into intermediate logical steps, achieving state-of-the-art accuracy on competitive mathematics (AIME), advanced programming challenges, and doctoral-level scientific benchmarks (GPQA).
Image inputs are supported alongside text, enabling visual reasoning over diagrams and charts. OpenAI o1 is hosted exclusively on OpenAI Direct API with no open-weight distribution or local runtime support.
Technical Specifications
Context Window
200,000 tokens
Max Input Tokens
200,000
Max Output Tokens
100,000
Architecture
Reasoning Transformer
Released
2024-12-05
Modalities
Text → Text (Both)
Capabilities
Text Generation
Code Generation
Structured Output
Reasoning
Function Calling
Long Context
Use Cases
Coding Assistance
Data Analysis
Research
Languages
English (en)
Spanish (es)
French (fr)
German (de)
Chinese (zh)
Japanese (ja)
Access Routes
| Route | Type | Protocol | Region / Scope | Provider | Status |
|---|---|---|---|---|---|
| OpenAI API | Hosted API (Direct provider) | OpenAI-compatible REST API | Global | OpenAI | Available |
Provider
Identifiers
- o1 Canonical identifier Default
- o1-2024-12-17 Canonical identifier Default
- o1-preview Alias