To add insult to injury, the cost of developing AI-powered applications will be drastically reduced for developers.
The flagship large language model that OpenAI uses to power ChatGPT was recently updated with a more compact and cost-effective version, which was released on Thursday. When compared to the GPT-3.5 Turbo, which was Open’s smallest model up until today, the new model, which is named GPT-4o tiny, will be sixty percent less expensive for developers to use in order to build AI-powered applications and services. Consumers, however, are the focus of this piece of news. Beginning today, the GPT-4o small will take the place of the GPT-3.5 Turbo for free users of ChatGPT. This implies that your experience with ChatGPT will greatly improve from its current state.
According to OpenAI, the GPT-4o mini was able to attain a score of 82 percent on a benchmark for the industry known as the MMLU. The acronym MMLU stands for “Measuring Massive Multitask Language Understanding,” and it consists of 16,000 multiple-choice questions that cover 57 different academic areas. Compared to the GPT-4o, which received a score of 88.7 percent, this value is little lower, but it is greater than the GPT-3.5 Turbo, which received just 70 percent. AI specialists have cautioned against depending on benchmarks of this kind to evaluate the level of intelligence possessed by AI systems; yet, up until this point, benchmarks have been the only method available for evaluating the effectiveness of big language models.
When it comes to the construction of AI-powered applications, builders have more leeway to work with when they have smaller versions of flagship models. It is important to remember that not everyone need or is able to afford access to the most powerful and largest models for every use case. Google stated in May that it had developed its very own lightweight variant, which it called Gemini 1.5 Flash. The business claimed that it was designed for speed and efficiency. Additional artificial intelligence businesses, such as Anthropic, also have miniature copies of their full-scale models.
Additionally, the GPT-4o small will soon be able to handle other sorts of content, such as audio and video, in addition to the text and images that it is currently capable of taking in and producing. And despite the fact that GPT-3.5 Turbo will no longer be available through ChatGPT, developers will still be able to access it using OpenAI’s application programming interface (API) in order to construct applications and services using it for the time being.