OpenAI has introduced a new model called "o1," part of a planned series of "reasoning" models designed to solve complex problems faster than humans. Alongside o1, OpenAI is also releasing a smaller, more affordable version called "o1-mini." This model, widely rumored as the "Strawberry model," represents a significant shift toward OpenAI's goal of human-like AI. O1 excels at coding and solving multistep problems, outperforming previous models, but it's more expensive and slower than GPT-4o. It's available as a "preview" to certain ChatGPT users, with developer access priced steeply at $15 per million input tokens and $60 per million output tokens—far more than GPT-4o.
The o1 model uses reinforcement learning and a "chain of thought" process, allowing it to tackle complex problems, like advanced math and programming, more effectively. OpenAI claims it hallucinates less but hasn’t fully eliminated the issue. The model significantly outperforms GPT-4o in areas like solving International Mathematics Olympiad problems but struggles with factual knowledge and lacks the ability to browse the web or process images and files.
OpenAI emphasizes that o1 is still in early stages, with its reasoning capabilities being a foundational step toward more advanced autonomous AI systems capable of decision-making. Despite its promise, o1 is currently slow and expensive for developers to utilize fully, but it represents progress toward human-like intelligence.
