OpenAI has just unveiled its latest AI models, o1-preview and o1-mini, marking a significant leap in artificial intelligence technology. These new models boast advanced reasoning capabilities, allowing them to “think” before responding to complex queries.
Key points about the new o1 series:
- OpenAI has released two models: o1-preview and o1-mini
- The models feature improved reasoning abilities and can demonstrate their thought process
- They’re designed for tackling complex tasks rather than everyday use
The o1 series represents a fresh start for OpenAI, with the naming convention reset to “1” to distinguish it from the company’s previous offerings. While these models are more resource-intensive and slower to respond compared to their predecessors, they offer unprecedented problem-solving abilities in fields like science, coding, and mathematics.
What sets o1 apart:
- Trained using a new optimization algorithm and dataset
- Employs reinforcement learning with a reward-punishment system
- Utilizes “chain of thought” prompting for problem-solving
- Demonstrates significantly improved accuracy in complex tasks
- Presents its reasoning process to users
The o1 model has shown remarkable performance, scoring 83% on the International Mathematics Olympiad (IMO), a substantial improvement over GPT-4o’s 13% score. However, there are limitations to its “thinking time,” and the model may indicate when it’s approaching this limit.
o1-mini, while more limited in scope, excels at coding tasks and is considerably more cost-effective than its larger counterpart. It’s particularly suited for tasks that don’t require extensive general knowledge.
Accessibility and usage:
- ChatGPT Plus and Team users can access both models immediately
- Educational and Enterprise users will gain access starting next week
- Free users currently don’t have access, but o1-mini may be available to them in the future
- Weekly message limits: 30 for o1, 50 for o1-mini
- API access is available for qualifying developers (Tier 5)
To use these models, ChatGPT users can manually select them from the menu. OpenAI is working on implementing automatic model selection based on the user’s prompt in future updates.
While the o1 series represents a significant advancement in AI technology, early user feedback suggests that it may still struggle with some basic questions. OpenAI is likely to continue refining these models to improve their performance across a broader range of tasks.
It’s worth noting that despite the introduction of the o1 series, OpenAI plans to continue developing and releasing new models in the GPT series, so we can still anticipate future advancements in that line as well.