OpenAI Expands Access to Advanced Reasoning Model for Free ChatGPT Users

OpenAI just made it easier to use its 'strawberry' reasoning model for free -- here's what's new

OpenAI has made o1-mini, the smallest version of its new reasoning family of artificial intelligence models, available for free in ChatGPT. You can even use it without signing up for an account. Unlike other AI models, such as GPT-4 and Anthropic's Claude, o1 can take a prompt and work through a problem step-by-step before responding with its final output. This significantly reduces error rates or "hallucinations" and allows for much longer, well-thought-out responses. There are some limitations to using the o1 family of models. It takes significantly longer to get a response, and there is no multi-modal capability, so you can't give it a document or an image. It only works with text. Giving it the name "mini" is also somewhat misleading. While it is a much smaller model than the o1-preview, in some respects, its performance is better as it was trained on a more modern architecture. As with the bigger GPT-4o, while o1-mini is available to free users, it does have rate limits, and you will find that you run out of them very quickly. In fact, these rate limits also apply to paid subscribers; those paying get a few more per week. When OpenAI first launched the o1 family of models last week, they said that users would only get 30 weekly messages to test it out. This is because it is a significantly more expensive model to run, as it goes over multiple prompts repeatedly until it gets to the response that it feels is the right one. Unlike other models, it is also able to generate significantly larger output, including entire codebases and multi-page reports, as we discovered during our early testing. OpenAI imposed limits on how often you can use it. Because of demand and people still trying to work out exactly what it's useful for, they have since significantly increased those limits over the initial 30 per week. First, OpenAI completely refreshed everyone's limits, so as of Friday, everyone has a completely clean slate. Then, the company announced it would be increasing the number of messages per week from 30 to 50 for o1-preview and to more than 200 for o1-mini (if you're a paid subscriber). If you are approaching one of the o1 family of models for the first time, you need to take a slightly different approach than you would with GPT-4o or Gemini. With those models, you tend almost to treat it like a conversation with Google: throw a question at it, add some extra details and hope for the best. However, you need to be more specific because the o1 family of models reasons through their responses. In one test, I was fairly vague and tasked it with creating a "gamified task manager app in iOS." Asking that question to ChatGPT would have resulted in all of the code to run the app, at least at a very basic level. However, o1-mini gave me a step-by-step design document for producing the app and an outline of the structure for the codebase. So, if you decide to use it to try and create an app, you must ensure that you're giving it specific and explicit instructions to provide the code necessary to develop the app. And that level of specificity translates across all use cases.

ChatGPT o1-mini reasoning model is already rolling out to free users - are you one of the lucky ones?

ChatGPT's new reasoning o1-mini AI model, previously codenamed 'Project Strawberry' arrived last week and it's already rolling out to free users. o1-mini and o1-preview are new AI models with impressive reasoning ability that can quickly solve riddles, answer equations, and even show you the thought process behind their decisions. If you want to see what o1-mini and o1-preview are capable of, check out Eric Hal Schwartz's test which showed the AI model was incredible at solving even the hardest riddles. At launch, OpenAI had promised free users would get to test out the reasoning model in the future, too. And that time has already come, with free users gaining access to o1-mini without subscribing to ChatGPT Plus. There is a caveat, however, as neither myself nor my colleague has access to o1-mini for free just yet, so it looks like you'll need to be one of the lucky ones to use 'Project Strawberry' for free right now. OpenAI hasn't confirmed how it's determining which free accounts to give access to but we expect it's a slow rollout that should grant everyone access in due course. Currently, o1-mini is a standalone AI model that you need to select before asking a complex question. To check if you've got access for free, simply log in to ChatGPT on your desktop and click 'ChatGPT Auto'. If you're lucky enough to have free access to o1-mini, you'll be able to select the AI model from the dropdown menu. I've been testing o1-mini over the last few days using my ChatGPT Plus subscription and I'm seriously impressed with the AI model's ability to solve the most complex of riddles. Internally, o1-mini has been used to solve puzzles like New York Times Connections and Wordle, a pretty impressive feat for artificial intelligence. There's no further info on the free version of o1-mini, such as how many prompts you can ask it per day, but just having access in general is a major step forward in improving ChatGPT for the masses, without signing up for a monthly fee.