GPT 4o Mini, Pricing, Limits and a Review
GPT-4o Mini is OpenAI’s new small and cost-efficient AI model. It packs the multimodal AI capabilities of GPT-4o into a much smaller model, which is now priced 3.33x lower then GPT-3.5 Turbo. Users should experience faster responses on the ChatGPT web or mobile app, and developers now have a much cheaper model to utilize in new use cases through the API.
Capabilities
GPT-4o Mini carries most of the same capabilities as GPT-4o, including its advanced reasoning capabilities and multimodal abilities (both vision with image inputs and text). The context window is the same as GPT-4o at 128k input tokens and 16k output tokens. It’s trained up to October 2023. The model is still highly capable even with its smaller size. OpenAI says, “GPT-4o Mini scores 82% on MMLU and currently outperforms GPT-4 on chat preferences in the LMSYS leaderboard.”
Pricing for Developers
GPT-4o Mini is included within the ChatGPT interface. For developers who want to build applications, the cost for the 4o Mini API is $0.15 per 1 million input tokens and $0.075 per 1 million output tokens at the time of launch on July 18, 2024. Batch pricing is also available with a 50% discount. This makes 4o mini roughly 33x cheaper than GPT-4 and 3x cheaper than GPT-3.5 Turbo.
New Use Cases
Being fast and cheap is going to unlock new use cases for developers. OpenAI says applications that require processing large context windows, such as inputting a lot of documents or code, will benefit from this new model. Additionally, applications that need to process many chained or parallel requests will benefit from the faster response times and low cost. OpenAI says, “We expect GPT-4o Mini will significantly expand the range of applications built with AI by making intelligence much more affordable.”
We tried 4o mini - It was fast and capable
When we tried GPT-4o Mini through the API, we found it was indeed able of processiing requests in parallel at a similar speed to 3.5 Turbo while providing results closer to GPT 4o level. You can access it today by using the “gpt-4o-mini” ID in API requests and it will be accessible through the official ChatGPT interface where the model is also available. We tried using it for code generation tasks and felt it's capability was comparable to 4o while speed and cost to GPT 3.5.