Tech Desk 
Jul 19, 2024
OpenAI, the leading artificial intelligence (AI) firm, has introduced a new, smaller and cheaper version of its GPT-4o generative large language model (LLM).
The new mini model is named GPT-4o Mini as a replacement for GPT-3.5 Turbo, the cheapest option for ChatGPT users to build applications using OpenAI technology.
It was first revealed in May during the OpenAI Spring Update, showing its capabilities such as speech-to-speech assistant and voice guide etc.
Read more: OpenAI for macOS fixes security concerns over plain text chat storage
The Mini model is a multimodal trained on material up to October 2023 that can handle more than just written text having a context window of 128,000 tokens and an ability to emit around 16,000 tokens of output.
OpenAI collaborated with companies like Ramp and Superhuman to check the model’s capabilities which found GPT-4o mini better than GPT-3.5 Turbo in performing tasks such as extracting structured data from receipts or generating high-quality email responses from thread history.
It’s important to note that the cost of Mini edition per million input and output tokens is quite less at 15 and 60 cents respectively, as compared to GPT-4o which costs $5 and $15 per million input and output tokens.
According to OpenAI, the GPT-4o Mini model is testing a new safety trick called Instruction Hierarchy that allows it to prioritise some instructions over others even those coming from OpenAI.
Despite being multimodal, GPT-4o Mini will be able to take in images and text but can only generate text. While other forms of content generation may be available in the future.
