OpenEye, a leading artificial intelligence laboratory, today releases a smaller, faster, and less expensive version of its highly successful AI model GPT-4o

The new mini model is designed to replace the GPT-35 Turbo This model was the default for free ChatGPT users and the least expensive option for developers building applications with OpenAI technology

GPT-4o was first released at the OpenAI Spring Update in May, showing its as-yet unreleased capabilities as a text-to-speech assistant and voice guide

The GPT-4o mini offers many of the same features as its big brother However, this model has fewer parameters and is smaller, making it much faster and cheaper, but without the overall reasoning and understanding of the 4o itself

ChatGPT free and paid users (including Teams plan users) will have access to the GPT-4o mini today, and it will be rolled out to Enterprise customers next week

OpenAI is the latest AI lab to offer a “baby” version of its flagship AI model It typically has the same core functionality but less overall performance than its big brother It is not clear if OpenAI is also working on GPT-4o, but GPT-4 appears to be fulfilling its function

Google has Gemini Nano and Pro for smaller versions and Ultra for the most complex tasks; Anthropic has Claude Haiku for simpler tasks, Sonnet for most use cases, and Opus for the most complex queries

Olivier Godement, OpenAI's API platform lead, told the Verge that mini is “working on OpenAI's mission to make AI more widely accessible to people”

Rate limits imposed on ChatGPT's free users meant that if the number of messages exceeded a certain number, they would be degrade to the two-year-old GPT-35, which has lower performance Now users can take advantage of the core features of GPT-4o

GPT-4o has not yet been unlocked in its full functionality by OpenAI due to safety concerns related to some multimodal features (such as the ability to mimic speech and even generate sound effects (as seen in some leaked versions))

According to OpenAI, the GPT-4o mini model is testing a new safeguard called the Instruction Hierarchy

GPT-4o mini can capture images and text, but despite being a multimodal model, it can only generate text; OpenAI says that other forms of content generation may be possible in the future