You Can Now Pay $200 a Month for ChatGPT “reasoning”
Would you pay $200 a month for unlimited ChatGPT? What if it can “reason”? OpenAI thinks you can do it.
As part of the 12 Days of Shipmas, where the company announces new features for 12 days in a row, OpenAI is finally bringing its first reasoning model out of preview, and also adding unlimited access to it and all OpenAI models for $200. monthly subscription plan.
The reasoning model, called OpenAI o1, has been available in preview since September . Paid ChatGPT members can send 30 messages per week to o1-preview and 50 messages per week to the lighter o1-mini. Now that it’s in full release, as CEO Sam Altman explained during today’s livestream , Plus and Team members will still be limited in how much they can use it (Enterprise and Edu members will also have to wait a week to receive access), but when they do it will presumably be much more powerful.
What is a reasoning AI model?
One of the biggest problems with AI is hallucinations , or when it simply does something wrong. Since an AI chatbot can only rely on its training, it usually cannot determine what is real and what is not, and will present lies with the same confidence as facts.
Reasoning AI is an attempt to correct this. Using a reasoning model, the AI breaks the clue into several parts, addressing each one in turn and doing its best to check the accuracy of its previous conclusions before moving on, while showing you its thought process. Additionally, the response may take longer than your typical model to prevent errors.
It’s called “chain of thought,” and in testing o1-preview and 01-mini , Lifehacker editor Jake Peterson had some luck with both simple clues (is a hot dog a sandwich?) and more complex ones (create a 6×6 nonogram puzzle that… solved looks like the letter Q). An early version of the bot took more than a minute to generate responses when needed and provided it with a drop-down menu to scroll through its “thought process.”
This ensured that both he and the bot could easily debug and understand where errors were coming from, and in the final o1 model, OpenAI promised that it had reduced “serious errors in complex real-world questions by 34%” and that the model was generally sound. is now “about 50% faster.”
In particular, OpenAI published charts promising that the new model is more than 50% more reliable than the non-reasoning GPT-4o model in coding and more than 40% more reliable in mathematical calculations. These are all internal numbers, and OpenAI hasn’t been entirely clear about how it tests or measures these models, but it’s a pretty big boast.
It will likely take some time for the experts to do their own independent testing, so perhaps you’ll soon see some cold water thrown at these claims. A recent Apple study , for example, found that o1’s “reasoning” abilities are still more like “sophisticated pattern matching.”
Would you pay $200 for ChatGPT?
This is where the rub comes in. OpenAI actually claims to have an improved version of o1, but it is very expensive. Along with OpenAI o1 was announced ChatGPT Pro, a new membership plan that gives unlimited access to all OpenAI models and also unlocks o1’s “pro mode”.
“O1 pro mode is judged by external test experts to provide more reliable and complete answers, especially in areas such as data science, programming and case law analysis,” OpenAI wrote in a blog post .
Essentially, Pro mode allows the model to use more calculations and spend more time, resulting in a slightly more than 10% increase in reliability depending on the task. Is a little extra performance worth it? Well, maybe if you’re a medical researcher or other power user, that’s probably why OpenAI is awarding 10 grants to “leading institutions in the US” that will give them free access to ChatGPT Pro.
Everyone else will have to decide how far they want to stretch their wallet, although OpenAI isn’t strictly aimed at enterprise customers here, and the live announcement says the o1 pro mode is also aimed at “power users” who are “already pushing models to their limits” in tasks such as mathematics, programming and writing.”
What does the future of ChatGPT look like?
While OpenAI o1 will likely be too expensive for most people at this point, even if they don’t consider its pro mode (ChatGPT Plus still costs $20 per month), the company has said it is committed to improving the model’s usability for “everyday options usage,” in addition to “really complex math and programming problems.” As part of today’s release, the model should now answer simple questions “very quickly” and take longer to answer more complex questions, rather than lingering on all queries.
In this way, OpenAI is paving the way for o1 to potentially replace its models without speculation in the future. This could be a big boon for free users, although it’s unlikely to happen anytime soon.
Meanwhile, sources told The Verge that Sora , OpenAI’s text-to-video model, will be released during the 12 Days of Shipmas event.