OpenAI Unveils o1: The New "Strawberry" Model with Advanced Thinking Abilities

OpenAI’s Project Strawberry, now officially called OpenAI o1, was fully unveiled on September 12. The new AI model is designed to "think before responding," enabling it to tackle complex tasks and solve more challenging problems in fields like science, coding, and math compared to earlier versions. This marks the debut of the o1 series within ChatGPT, with OpenAI planning regular updates and enhancements. However, these advanced capabilities come with a higher cost. Let’s take a closer look at its functionality.

OpenAI o1: What is it and how does it work?

OpenAI's o1 model represents a major leap forward in AI development. According to the company, it's a key step toward more human-like intelligence, showing noticeable improvements in generating code and handling intricate, multi-step challenges compared to its predecessors.

OpenAI explains that the model is trained to think through problems more thoroughly before responding, similar to how a human would. This allows it to refine its problem-solving process, experiment with different strategies, and learn from mistakes. The model shines in areas like math and coding, and future updates will reportedly see it perform at a level comparable to PhD students in fields like physics, chemistry, and biology. However, it’s not without limitations.

Limitations and Differences

Despite its advancements, OpenAI o1 lacks certain features found in the GPT-4o model, such as browsing the web for information or uploading files and images. Currently, it only accepts text inputs, so tasks like solving problems from uploaded photos aren’t supported yet.

In essence, OpenAI o1 is ideal for solving complex academic and mathematical problems that require deeper reasoning. But there’s more to know.

OpenAI CEO Sam Altman shared on X that while o1 shows promise, it’s still flawed and has limitations. He also shared graphs showcasing its performance in solving reasoning-based tasks, such as math and coding, and PhD-level scientific questions.

It’s also slower than current models. OpenAI attributes this to the fact that o1 is designed to "think before answering," which means it takes longer to process queries. The company notes that in many cases, GPT-4o may still be more effective in the short term.

OpenAI o1 Mini: A Cheaper, Faster Alternative

To provide a more affordable option, OpenAI also launched OpenAI o1-mini, a faster and cheaper reasoning model that excels in coding. This smaller model costs 80% less than o1-preview, making it a cost-efficient choice for tasks requiring reasoning without needing extensive world knowledge.

Who Can Access OpenAI o1?

Starting today, ChatGPT Plus and Team users can access both o1-preview and o1-mini in ChatGPT by selecting them manually from the model picker. At launch, usage is limited to 30 messages per week for o1-preview and 50 for o1-mini, but OpenAI is working to expand these limits and enable automatic model selection based on the prompt.

While OpenAI plans to extend o1-mini access to all free-tier ChatGPT users, o1 comes with a higher price tag. For API usage, o1-preview costs $15 per 1 million input tokens and $60 per 1 million output tokens. In comparison, GPT-4o is priced at $5 for 1 million input tokens and $2.50 for 1 million output tokens. In India, this translates to around ₹1,260 for 1 million input tokens with OpenAI o1, compared to ₹420 for GPT-4o.

The ChatGPT Plus subscription remains priced at $20 per month, roughly ₹1,650 in India.