The latest small AI model from OpenAI, GPT-4o mini, was released on Thursday. The company says that starting today, a cheaper and faster AI model called GPT-4o mini will be made available to developers as well as to customers through the ChatGPT web and mobile app. Next week, enterprise users will be able to get in.
The company says that GPT-4o mini does better than the best small AI models in the business at tasks that require reasoning with both text and images. As small AI models get better, developers like them more because they are faster and cheaper than bigger models like the GPT-4 Omni or the Claude 3.5 Sonnet. They are a good choice for developers who want to use AI models to do a lot of simple jobs over and over again.
The GPT-4o mini will take the place of the GPT-3.5 Turbo as OpenAI’s smallest model. The company says its newest AI model gets 82% on MMLU, a test for thinking. This is higher than the 79% scores for Gemini 1.5 Flash and the 75% scores for Claude 3 Haiku, according to data from Artificial Analysis. GPT-4o mini got an 87% on the MGSM test of math thinking, while Flash got a 78% and Haiku got a 72%.
OpenAI also says that the GPT-4o mini is much cheaper to run than its previous frontier models and more than 60% less expensive than the GPT-3.5 Turbo. The GPT-4o mini model can currently handle text and vision in the API. According to OpenAI, the model will soon be able to handle video and voice as well.
In a conversation with TechCrunch, Olivier Godement, head of Product API at OpenAI, said, “We need to make the models much cheaper so that AI can help everyone.” “I believe the GPT-4o mini is a huge step in the right direction.”
It costs 15 cents per million input tokens and 60 cents per million output tokens for coders who use OpenAI’s API. The model has a knowledge end date of October 2023 and a context window of 128,000 tokens, which is about the length of a book.
OpenAI wouldn’t say how big GPT-4o mini is, but they did say it’s about the same size as other small AI models like Llama 3 8b, Claude Haiku, and Gemini 1.5 Flash. Based on tests done before launch in the LMSYS.org chatbot arena, the company says GPT-4o mini is faster, cheaper, and smarter than the best small models on the market. Early tests done by different groups seem to prove this.
In an email to TechCrunch, George Cameron, Co-Founder at Artificial Analysis, said, “Compared to similar models, GPT-4o mini is very fast, with a median output speed of 202 tokens per second.” “This is more than twice as fast as GPT-4o and GPT-3.5 Turbo, and it’s a great option for use cases that need speed, like many consumer apps and agent-based ways to use LLMs.”
The new tools from OpenAI for ChatGPT Enterprise
On the same day, OpenAI also announced new tools for business buyers. OpenAI wrote a blog post about the Enterprise Compliance API, which will help companies in government, healthcare, finance, and other heavily controlled fields follow the rules for logging and auditing.
The company says that these tools will let users check their ChatGPT Enterprise data and do something with it. The API will store records of exchanges with timestamps, such as conversations, uploaded files, workspace users, and more.
Also Read: Openai Lets People See What Its Ai is Doing Behind the Scenes
OpenAI is also giving admins more power over workspace GPTs, which are a special kind of ChatGPT made for business needs. Before, admins could only fully allow or stop GPT actions that were made in their own workspace. Now, owners of workspaces can make a list of domains that GPTs are allowed to interact with.
What do you say about this story? Visit Parhlo World For more.
