The company OpenAI presented a new economic model of artificial intelligence called GPT-4o mini, which allows to reduce the cost of using artificial intelligence and expand its application. This model performed well in comprehension and text analysis tests, reaching 82% on the MMLU. This means that the GPT-4o mini is able to effectively understand and interpret text information.

One of the key advantages of the GPT-4o mini is its affordability. The cost of information processing is much lower than that of previous models. OpenAI measures the amount of information in tokens that roughly correspond to words or parts of words. Processing a million incoming tokens costs 15 cents (about 6 hryvnias), and a million outgoing tokens costs 60 cents (about 25 hryvnias). This makes the GPT-4o mini an attractive alternative for those looking for cost-effective AI solutions.

The GPT-4o mini supports low-cost, low-latency tasks such as processing large amounts of context, chaining or parallelizing model calls (e.g. using multiple APIs), or interacting with customers via fast, real-time text responses (e.g. chatbots customer support). The model also supports text and images in the API, and will support video and audio inputs and outputs in the future. It has a context window of 128K tokens, supports up to 16K output tokens on demand, and has knowledge until October 2023.

An improved tokenizer makes processing text in other languages ​​more cost-effective. In ChatGPT, users of Free, Plus and Team tariff plans can now access GPT-4o mini instead of GPT-3.5. In May, OpenAI introduced another new AI model, GPT-4o, which is a voice assistant capable of discussing audio, images, and text in real-time.

The GPT-4o responds to audio input in 232 milliseconds, which corresponds to the speed of a human conversation. With this model, users can interact with ChatGPT in a conversational manner, even interrupting it in the middle of a response, and the model demonstrates real-time adaptability. GPT-4o is also able to recognize emotional nuances in the voice and respond in different emotional styles.

The GPT-4o is said to be able to better perceive images, audio and other visual information compared to previous models, greatly improving its speech processing capabilities. New features include uploading screenshots and asking questions based on them, remembering previous questions, and the ability to go online directly.