On Thursday, OpenAI unveiled its latest AI innovation, the GPT-4o mini, a compact AI solution designed to be more economical and efficient than its predecessors. This new model, accessible via GPT-4o mini, is aimed at developers and is also available through the ChatGPT web and mobile applications for general users starting from today. Access for enterprise clients will be rolled out in the following week.
According to OpenAI, the GPT-4o mini surpasses the performance of leading compact AI models on textual and visual reasoning tasks. The growing preference for small AI models among developers is attributed to their quick processing times and reduced operational costs, especially when compared to larger variants like the GPT-4 Omni or Claude 3.5 Sonnet. These models offer an efficient solution for handling large volumes of straightforward tasks.
Replacing GPT-3.5 Turbo, the GPT-4o mini emerges as OpenAI’s smallest offering. It boasts an 82% achievement rate on the MMLU benchmark, a critical measure of reasoning abilities, outperforming Gemini 1.5 Flash and Claude 3 Haiku with scores of 79% and 75% respectively. This data is supported by Artificial Analysis. In mathematics reasoning on the MGSM benchmark, GPT-4o mini scored an impressive 87%, surpassing others significantly.
Moreover, running GPT-4o mini is significantly more cost-effective than previous models, showing over a 60% reduction in cost against the GPT-3.5 Turbo. Presently supporting both text and vision capabilities, OpenAI anticipates future upgrades to include video and audio functionalities.
Olivier Godement, the Head of Product API at OpenAI, shared with TechCrunch, “To ensure AI’s empowerment across globally diverse regions, affordability is key. GPT-4o mini represents a significant leap towards achieving this goal.”
For those integrating OpenAI’s API into their projects, GPT-4o mini comes at an attractive price of 15 cents per million input tokens and 60 cents for the same amount of output tokens. It features a sizable context window of 128,000 tokens, equivalent to the length of a book, and is up-to-date as of October 2023.
The exact dimensions of GPT-4o mini remain undisclosed by OpenAI. Nevertheless, it’s categorized along with other compact models such as Llama 3 8b, Claude Haiku, and Gemini 1.5 Flash. OpenAI touts its latest model for its superior speed, cost efficiency, and intelligence, as evidenced in pre-launch evaluations within the LMSYS.org chatbot environment. Initial third-party tests appear to corroborate these claims.
“When compared to its counterparts, GPT-4o mini stands out for its remarkable speed, managing a median of 202 tokens per second,” stated George Cameron, Co-Founder of Artificial Analysis, in a communication to TechCrunch. “This speed, over two times faster than that of GPT-4o and GPT-3.5 Turbo, offers significant advantages for time-sensitive applications and various consumer-oriented and assistant-based implementations of LLMs (Large Language Models).”
In a separate development, OpenAI introduced new toolsets for enterprise customers. As mentioned in a recent publication, these tools include an Enterprise Compliance API, designed to aid businesses in navigating the complex compliance landscape in sectors like finance, healthcare, legal services, and government by providing detailed records of user interactions and more.
These innovative tools enable administrators to conduct audits and manage data within the ChatGPT Enterprise ecosystem effectively, offering a detailed log of interactions that includes timestamps, conversation histories, uploaded documents, and user activities, among others. Moreover, administrators are given increased flexibility in managing interactions within their GPTs, allowing the creation of approved domain lists for interactions, a significant enhancement over the previous all-or-nothing access control scheme.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
