o1

Basic Information

OpenAI o1 is the next-generation reasoning model launched by OpenAI. As the successor to o1-preview, its first official version is o1‑2024-12-17. This version has been fine-tuned and optimized based on the model released in ChatGPT two weeks ago, and is now gradually being made available to developers at API usage tier 5, with production-grade availability. It is specifically designed to handle complex, multi-step tasks; previously, developers have leveraged its preview version to build intelligent agent applications such as smart customer service systems, optimized supply chain decisions, and predicted financial trends.

Core Features

o1 integrates several practical features: it supports function calls and can seamlessly connect to external data and APIs; thanks to its structured output capability, it can strictly adhere to custom JSON schemas when generating responses; developers can define the model’s tone, style, and other behavioral guidelines by setting developer messages; its visual capabilities enable image-based reasoning, expanding its application scenarios in fields such as science, manufacturing, and programming. Additionally, the model introduces a new reasoning_effort API parameter, allowing developers to control the model’s thinking time and better match its reasoning needs for different tasks.

Technical Highlights

In terms of performance, o1‑2024-12-17 has set new records in multiple benchmark tests: it achieved a GPQA diamond score of 75.7, a MATH (pass @1) score of 96.4, and a SWE-bench Verified score improved to 48.9. Moreover, it significantly outperformed GPT-4o in both function call and structured output tests. In terms of efficiency, when processing the same requests, o1 uses an average of 60% fewer inference tokens than o1-preview, with lower latency, while maintaining high inference accuracy and optimizing resource consumption.

Market Impact

The launch of o1 further strengthens OpenAI’s leading position in the field of AI reasoning models, providing developers with more powerful tools to build highly complex intelligent agent applications. It is expected to drive upgrades in AI applications across industries such as finance, manufacturing, and customer service. Its phased access strategy and ongoing performance optimizations will also attract more developers to innovate based on this model, accelerating the deployment of AI-native applications and prompting the industry to explore further research and application of high-precision reasoning models.