OpenAI adds new features and updates to its flagship product, ChatGPT, on a fairly regular basis. Usually these are minor changes, including adjustments to how a model responds. But last week the company released a completely new version of GPT-4oβand told no one about it.
I reported last week that ChatGPT had a “secret update” in that its replies felt different, faster, and more responsive than before, but that was just anecdotal. Now, there seems to have been a change.
OpenAI stated via its ChatGPT app account on X: “As of last week, there is a new GPT-4o model in ChatGPT,” adding the message: “I hope you all like it. Try it out if you haven’t already! We think you’ll like it.”
In typical OpenAI fashion, details on the new model are as sparse as the announcement. All we know is that it is not “gpt-4o-2024-08-06,” the new API model released last week to make it cheaper for developers to access the power of GPT-4o. If you don’t want to use ChatGPT, I’ve tried some good alternatives.
What’s new in ChatGPT?
as of last week there is a new GPT-4o model in ChatGPT. I hope you all like it and try it out if you haven’t already! We think you will like it π12 August 2024
OpenAI hasn’t said anything official other than the X-post, which has sparked wild speculation on social media. Some believe this is an upgrade from GPT-4o Large, a supposedly larger version of GPT-4o with more parameters than GPT-4o or GPT-4o mini.
This theory doesn’t convince me because if it was a completely new model, we would probably have seen an announcement like we did for GPT-4o mini. However, it is logical that OpenAI follows other AI labs like Anthropic and Google and releases models of different sizes.
OpenAI also tends to test new models in the chatbot space of LMSys, where humans blindly and directly compare two models. There was a popular anonymous chatbot in this space, which many now suspect is the update to GPT-4o.
Whether it’s a fine-tuned version of GPT-4o, a new larger model, or even the first sighting of Project Strawberry, what we have is a better ChatGPT that refuses to respond less often.
How do I access the new model?
OpenAI says the new model simply replaced the current version of GPT-4o in ChatGPT, so accessing it is easy: visit the ChatGPT website or open the app and chat.
It’s not clear if the update also applies to GPT-4o mini, but I suspect this is just the full GPT-4o since OpenAI wrote on X that it’s available to free users with a message limit.
You’ll need a $20-a-month ChatGPT Plus subscription to push the model to its limits, but you get enough messages per day to try it out for free. If you run out of messages and don’t want to upgrade, you can simply upgrade to GPT-4o mini, a fine-tuned smaller version of GPT-4o.