OpenAI surprised the tech world last week with the release of Project Strawberry, now known as “o1.” This new AI model series represents a significant leap forward, focusing on enhanced reasoning capabilities. While analysts predicted a later release, o1-preview and its lighter counterpart, o1-mini, are now available for evaluation and use. Let’s explore how to access this groundbreaking technology.
OpenAI CEO Sam Altman tweeted about the release, highlighting the model’s capabilities and limitations:
here is o1, a series of our most capable and aligned models yet:
o1 is still flawed, still limited, and it still seems more impressive on first use than it does after you spend more time with it. pic.twitter.com/Qs1HoSDOz1
— Sam Altman (@sama) September 12, 2024
Understanding OpenAI’s o1
OpenAI’s pursuit of Artificial General Intelligence (AGI) is well-documented, and o1 marks a significant stride toward this ambition. As the first in a new line of “reasoning” models, o1 is designed to deliberate before responding, a key differentiator from previous models. This allows o1 to tackle complex tasks and solve more challenging problems in science, coding, and mathematics.
OpenAI claims o1 mimics human reasoning, refining its thought processes, exploring different strategies, and learning from mistakes through training. Its performance is reportedly on par with Ph.D. students in physics, chemistry, and biology, achieving comparable results on benchmark tests. Furthermore, o1 excels in coding and mathematics, scoring 83% on an International Mathematics Olympiad (IMO) qualifying exam—a stark contrast to GPT-4o’s 13%—and reaching the 89th percentile in a Codeforces competition against human programmers.
o1-mini: A Lightweight Powerhouse
o1-mini, a streamlined version of o1-preview, offers significant cost savings, operating at 80% less expense than its larger counterpart. This makes o1-mini particularly efficient for coding analysis and generation tasks.
Accessing o1-preview
o1-preview models launched on September 12th for ChatGPT Plus and Teams subscribers, with Enterprise and Educational access following shortly after. Currently, access is exclusive to paying subscribers.
Enhanced Safety and Security
OpenAI emphasizes o1’s improved safety features. A new safety training program leverages the model’s advanced reasoning capabilities to better adhere to safety and alignment guidelines. In tests resisting jailbreak attempts, o1 scored an impressive 84 out of 100, significantly outperforming GPT-4o’s score of 22.
How to Subscribe and Use o1-preview
To experience o1-preview, a $20/month ChatGPT Plus subscription is required. Upgrade your plan through the left-hand navigation pane and follow the payment prompts. Once subscribed, select either o1-preview or o1-mini from the model picker on the ChatGPT homepage.
Note that usage is limited, even for subscribers, with a weekly cap of 30 messages for o1-preview and 50 for o1-mini. While OpenAI plans to offer free-tier access to o1-mini eventually, no official date has been announced.
