o1 Models by OpenAI
About
OpenAI's o1 family of large language models (LLMs) represents a significant advancement in AI's reasoning capabilities 4712. Unlike previous models that primarily focused on predicting the next word in a sequence, o1 models are trained to "think" before responding, employing a chain-of-thought process to solve complex problems 4712. This involves breaking down complex tasks into smaller, manageable steps, exploring different approaches, and correcting errors along the way 4712. The o1 models demonstrate improved performance on various benchmarks, including those involving mathematics, coding, and science, rivaling or exceeding the capabilities of human experts in certain areas 4712. However, these models are still under development and may lack some features found in earlier models, such as web browsing and image processing 15.
Current Variants
Use-when guidance is derived from seed capabilities, context, release, and replacement fields.
Use when the workload needs 200k context and structured outputs.
Use when the workload needs 128k context, reasoning, and code execution.
Use when the workload needs 128k context, reasoning, and code execution.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| o1-pro | Use when the workload needs 200k context and structured outputs. | 2024-12 | 200k contextstructured outputs | Current |
| o1-preview (09-12) | Use when the workload needs 128k context, reasoning, and code execution. | 2024-09 | 128k contextreasoningcode execution | Current |
| o1-mini (09-12) | Use when the workload needs 128k context, reasoning, and code execution. | 2024-09 | 128k contextreasoningcode execution | Current |
Release Timeline
2 release groupsReplaced By
Keep for legacy integrations; evaluate o3 before new work.
Specifications(4 models)
| Model | Released | Context | Reasoning | Structured Outputs | Code Exec |
|---|---|---|---|---|---|
| o1-pro | 2024-12 | 200k | No | Yes | No |
| o1-preview (09-12) | 2024-09 | 128k | Yes | No | Yes |
| o1-mini (09-12) | 2024-09 | 128k | Yes | No | Yes |
Available From(3 providers)
Pricing
| Model | Provider | Input / 1M | Output / 1M | Type |
|---|---|---|---|---|
| o1-mini (09-12) | Replicate API | $1.1 | $4.4 | Serverless |
| o1-pro | OpenRouter | $150 | $600 | Serverless |
Comparisons
- o1 (12-17) vs Claude 3.7 Sonnet
- o1 (12-17) vs DeepSeek R1
- GPT-4o (08-06) vs o1 (12-17)
- o3 vs o1 (12-17)
Frequently Asked Questions
- What is o1 used for?
- o1 is used for reasoning, structured outputs, and code execution. The family description and listed model capabilities point to those workloads as the best fit.
- How does o1 compare to GPT Realtime 2?
- o1 by OpenAI is strongest where you need reasoning, while GPT Realtime 2 by OpenAI is the closest related family to check for translation. o1 has 4 listed variants and reaches up to 200k context, while GPT Realtime 2 reaches up to 131k context, so compare the specs and pricing tables before choosing a production model.
- Which o1 model should I use?
- For the lowest listed input price, start with o1-mini (09-12) through Replicate API at $1.1/1M input tokens. For the most capable/latest local choice, evaluate o1-preview (09-12) with 128k context and reasoning.
