If you’ve somehow avoided using ChatGPT until now, there’s a new reason to try it out. OpenAI launched a new lightweight AI model primarily targeting developers looking to deploy AI features in their apps without breaking the bank. GPT-4o mini is cheaper than any other ChatGPT model so far, and it’s supposed to be faster than similar alternatives from OpenAI’s biggest rivals.
However, GPT-4o mini isn’t just for developers. The new model is available to all ChatGPT users, including people who are still on the ChatGPT Free subscription. It replaces the old GPT-3.5 Turbo LLM, and it should bring better performance than GPT-3.5 and a much better knowledge cutoff.
Here, I’ll explain everything you need to know about ChatGPT GPT-4o mini.
ChatGPT Free users can already access GPT-4o for free. That’s OpenAI’s best model, announced in early May just before Google unveiled all its AI innovations at I/O 2024. However, free ChatGPT accounts have limited access to GPT-4o. That’s where GPT-4o mini can come in handy. It can replace GPT-4o if and when you reach your quota on 4o.
OpenAI explained all the novelties the lightweight language model can deliver in a blog post. First of all, it is 60% cheaper than GPT-3.5 Turbo. It costs $0.15 per million input tokens and $0.60 per million output tokens. This detail is for developers, however. It doesn’t mean anything to ChatGPT Free users who won’t have to pay for access.
What does matter are the actual features. GPT-4o mini includes support for text and vision in the API and in ChatGPT apps. OpenAI promises better multimodal functionality in the future. That will include text, image, video, and audio inputs.
Moreover, GPT-4o mini has a context window of up to 128K tokens and support for up to 16K output tokens per request. However, ChatGPT Free users get an 8K token context window. If you’re on ChatGPT Plus or Teams, you’ll get context windows of up to 32K tokens. The Enterprise tier receives the full 128K token context window upgrade.
Finally, GPT-4o mini comes with a knowledge cutoff of October 2023, which is far better than the September 2021 cutoff of GPT-3.5 Turbo.
ChatGPT Free users should also know they have unlimited access to GPT-4o mini compared to GPT-4o. However, ChatGPT Free might not be as fast as the premium tiers.
As for performance, OpenAI’s blog has plenty of benchmark scores that show GPT-4o mini can outperform its direct rivals, Gemini Flash and Claude Haiku. According to the company, GPT-4o mini outperformed competitors in reasoning tasks, math and coding proficiency, and multimodal reasoning.
OpenAI also tackled safety and security in its GPT-4o mini blog post, saying the new model has the same safety mitigations built-in as GPT-4o. However, it also features improvements that might prevent abuse, such as ChatGPT jailbreak attempts:
Building on these learnings, our teams also worked to improve the safety of GPT-4o mini using new techniques informed by our research. GPT-4o mini in the API is the first model to apply our instruction hierarchy(opens in a new window) method, which helps to improve the model’s ability to resist jailbreaks, prompt injections, and system prompt extractions. This makes the model’s responses more reliable and helps make it safer to use in applications at scale.
As for reliability, this is still a genAI chatbot. You should expect errors, or “hallucinations,” from GPT-4o mini. I just ran into a few while briefly testing the model. I’m still on the ChatGPT Plus plan rather than the Free tier. Needless to say, errors can be expected regardless of whether you pay for ChatGPT.
I wouldn’t be surprised if GPT-4o mini becomes the default ChatGPT model for Apple Intelligence’s ChatGPT functionality once iOS 18 rolls out. However, that’s speculation at this point. For the time being, ChatGPT Free and Plus users have access to GPT-4o and GPT-4o mini in the ChatGPT app. It’s up to you to choose which model to use.
