OpenAI's official usage statistics or reports

OpenAI Unveils o1 AI Model with Advanced Reasoning, Fact-Checking, and PhD-Level Performance

Sep 12, 2024, 05:28 PM

OpenAI has announced the release of its new AI model series, OpenAI o1, previously codenamed 'Strawberry'. The o1 model is designed to perform complex reasoning tasks, including solving math, science, and coding problems. It employs reinforcement learning to think through problems before responding, a method claimed to improve accuracy and performance. The o1 series includes a smaller, cost-effective version called o1-mini, which is 80% cheaper and still highly effective at coding tasks. The models are available immediately for ChatGPT Plus and API users. OpenAI claims that o1 rivals human experts in reasoning-heavy benchmarks, can fact-check itself, and has demonstrated significant improvements over previous models, such as GPT-4o. Remarkably, the o1 model can also code an entire video game from a single prompt and performs at PhD-level in various academic tasks.