Computing ultimately comes down to 1s and 0s. For example: if you tack an extra 0 onto the $20 price tag for ChatGPT Plus, you’ll have the cost of ChatGPT Pro—the newest premium tier of OpenAI’s chatbot platform that offers access to the company’s new “reasoning” model.
The new subscription option, marketed for engineers and researchers, will cost $200 per month and offers unlimited use of the company’s GPT-4o and o1 models, as well as full access to the o1 model’s “pro mode”, which is designed to mimic human reasoning to answer more complex questions. The announcement was made as part of OpenAI’s “12 Days of Shipmas” during which it plans to show off 12 new products in the lead-up to the holiday season.
OpenAI previewed o1 earlier this year with limited access to its o1-preview and o1-mini models, previously known by the codename Strawberry. The next-generation large language models showed off the company’s new approach to complex computations through chain-of-thought reasoning. Basically, it is a chatbot capable of “thinking” before it responds to questions. Whereas chatbots operating on models like GPT-4o or GPT-4 might require refining prompts and questions in order to deliver a meaningful answer, o1 is designed to do all that work behind the scenes before responding.
The results, according to the company’s benchmarking tests, are impressive. OpenAI claims the o1 model scored in the 89th percentile in programming competitions held by Codeforces and was able to correctly answer 83 percent of questions from an International Mathematics Olympiad qualifying test. By contrast, GPT-4o only managed to get 14 percent right.
But the model has its fair share of shortcomings, too. It’s both slower and more expensive than GPT-4o and other models. In the preview version of o1, input tokens—essentially units of text that the model uses to parse a prompt—cost about three times the price of tokens for GPT-4o. An analysis of o1 conducted by the AI developer platform Vellum found the reasoning model is 30 times slower than its predecessor.
Reviews of the model also found that while it does seem more capable of tackling complex math problems and coding tasks, it’s no better—and in some cases, worse—at answering simple questions. In fact, OpenAI’s own help pages admit, “GPT-4o is still the best option for most prompts.”
Of course, the new ChatGPT Pro subscription gives users access to both the GPT-4o and new o1 reasoning model, so you could always switch between options depending on your needs. Or, alternatively, you could spend $200 on anything else.
+ There are no comments
Add yours