Exploring the Boundaries of Artificial Intelligence

Breakthrough research from language models to multimodal intelligence

We are committed to advancing the development of artificial intelligence technology, sharing our discoveries with the global research community through publishing papers, releasing models, and open-sourcing tools.

Core Research Areas

Our research covers key technology areas on the path to Artificial General Intelligence (AGI).

Generative Models

We developed the GPT series of language models, demonstrating the enormous potential of large-scale unsupervised learning in natural language understanding and generation. From GPT-1 to GPT-4, the models' capabilities continue to improve, showing amazing reasoning and creativity.

Computer Vision

Through the DALL-E series and CLIP, we explored the connection between text and images. DALL-E 3 can generate extremely realistic images from natural language descriptions, while Sora extended this capability to video generation.

Alignment Research

We are committed to solving the AI alignment problem, ensuring that AI systems' goals and behaviors remain consistent with human values. Through techniques like RLHF (Reinforcement Learning from Human Feedback), we have significantly improved model safety and utility.

Reasoning & Planning

We are researching how to give AI models stronger logical reasoning, mathematical problem-solving, and long-term planning capabilities. The OpenAI o1 series models have achieved breakthrough progress in complex reasoning tasks.

Breakthrough Achievements

GPT-4

GPT-4

GPT-4 is our most advanced system, capable of solving difficult problems with greater accuracy. It has broad general knowledge and problem-solving abilities, demonstrating human-level performance on various professional and academic benchmarks.

  • Process text over 25,000 words
  • Accept images as input and generate descriptions, classifications, and analysis
  • Advanced reasoning capabilities surpass previous models
DALL-E 3

DALL·E 3

DALL·E 3's ability to understand nuances and details is significantly better than previous systems, allowing you to easily transform ideas into highly accurate images. It is natively integrated with ChatGPT to help you generate prompts.

  • Extremely high instruction following ability
  • Generate realistic images and artworks
  • Built-in safety mitigation measures
Sora

Sora

Sora is our text-to-video model. It can generate videos up to one minute long while maintaining visual quality and adhering to user prompts. Sora can generate complex scenes containing multiple characters, specific types of motion, and accurate details of subjects and backgrounds.

  • Generate high-definition videos up to 60 seconds
  • Understand the laws of motion in the physical world
  • Support extension from images or existing videos