Model Overview Card for OpenAI o1-preview

Basic Information

• Model Name: OpenAI o1-preview
• Developer/Creator: OpenAI
• Release Date: September 12, 2024
• Version: o1-preview-2024-09-12
• Model Type: Large Language Model (LLM)

Description

Overview

OpenAI o1-preview is an advanced language model designed for complex reasoning and problem-solving tasks, particularly excelling in science, coding, and mathematics.

Key Features

• Chain-of-Thought (CoT) reasoning capabilities
• Enhanced performance in coding and mathematical tasks
• Self-fact-checking abilities
• Improved safety measures and alignment

Intended Use

The o1-preview model is intended for applications requiring deep reasoning and can accommodate longer response times. It's particularly useful for:

• Complex code generation and analysis
• Advanced mathematical problem-solving
• Comprehensive brainstorming sessions
• Multifaceted document comparison

Language Support

While specific language support details are not explicitly mentioned, the model demonstrates strong performance across various languages, including low-resource languages.

Context Window

The context window size is 128000 tokens.

Max Output Tokens

The maximum output token limit for o1-preview is 32,768 tokens.

Beta Limitations

During the beta phase, many chat completion API parameters are not yet available. Most notably:

• Modalities: text only, images are not supported.
• Message types: user and assistant messages only, system messages are not supported.
• Streaming: not supported.
• Tools: tools, function calling, and response format parameters are not supported.
• Logprobs: not supported.
• Other: temperature, top_p and n are fixed at 1, while presence_penalty and frequency_penalty are fixed at 0.
• Assistants and Batch: these models are not supported in the Assistants API or Batch API.

Technical Details

Architecture

The o1-preview model utilizes a transformer-based architecture with significant enhancements in reasoning capabilities. It employs large-scale reinforcement learning to perform chain-of-thought reasoning.

Training Data

• Data Source and Size: Trained on a vast dataset up to October 2023
• Knowledge Cutoff: October 2023

Performance Metrics

• Achieved 83% accuracy on a qualifying exam for the International Mathematics Olympiad
• Scored in the 89th percentile on Codeforces (Competitive Programming)
• Surpassed human PhD-level performance on the GPQA diamond benchmark in physics, chemistry, and biology

Comparison to Other Models

• Accuracy: Outperforms GPT-4o on most reasoning-heavy tasks
• Speed: Slower than previous models like GPT-4o, as it "thinks before it answers"
• Robustness: Demonstrates improved performance with strategic test-time submissions

Usage

Code Samples

API Documentation

For detailed API documentation, visit: https://docs.aimlapi.com/quickstart/setting-up

Reasoning models https://platform.openai.com/docs/guides/reasoning

Ethical Guidelines

OpenAI has implemented enhanced safety measures for the o1-preview model:

• Improved performance on internal safety tests compared to GPT-4o
• New safety training approach harnessing reasoning capabilities
• Collaboration with U.S. and U.K. AI Safety Institutes for evaluation and testing

Price

The o1-preview model is available through AI/ML API services. Pricing is set at $0.01575 per 1K input tokens and $0.063 per 1 K output tokens.