OpenAI has introduced a lighter and cheaper model for developers called GPT-4o Mini. This model costs significantly less than the full versions, and is considered more capable than GPT-3.5.
Using the OpenAI model to build applications can be very expensive. Developers who don’t have enough resources often choose cheaper models like Google Gemini 1.5 Flash or Anthropic Claude 3 Haiku. Now, OpenAI is entering the market with lighter models.
“I think the GPT-4o Mini really hits on OpenAI’s mission to make AI more accessible to a wider range of people. If we want AI to benefit every part of the world, every industry, every application, we have to make AI much more accessible,” Olivier Godement, API Product Platform Lead, told The Verge.
Starting today, ChatGPT users on Free, Plus and Team plans can use GPT-4o Mini instead of GPT-3.5 Turbo, while Enterprise plan users will get access next week. This means that GPT-3.5 will no longer be an option for ChatGPT users, but will still be available for developers via the API until they switch to GPT-4o Mini. Godement said that GPT-3.5 will be retired from the API at some point, but it is not yet certain when.
The new, lighter model will also support text and vision in the API, and the company says it will soon be able to handle all multimodal inputs and outputs like video and audio. With all these capabilities, this could lead to more advanced virtual assistants that can understand your travel plan and make suggestions. However, the model is intended for simple tasks, so no one is planning to build Siri for cheap.
This new model scored 82 percent on the Measuring Massive Multitask Language Understanding (MMLU) test, which consists of about 16,000 multiple-choice questions from 57 academic subjects. When the MMLU was first introduced in 2020, most of the models were pretty bad at it, which was the point because the models had become too advanced for the previous tests. The GPT-3.5 scored 70 percent on this test, the GPT-4 scored 88.7 percent, and Google claims the Gemini Ultra’s highest ever score of 90 percent. In comparison, competing models Claude 3 Haiku and Gemini 1.5 Flash scored 75.2 percent and 78.9 percent, respectively.
It’s worth noting that researchers are skeptical of tests like the MMLU, because the way they’re administered varies from company to company. This makes it difficult to compare different models, as reported by The New York Times. There is also the problem that the AI may already have the answers in its dataset, allowing it to “cheat”, and there is usually no third party evaluating the process.
For developers looking to build AI applications at a low cost, the launch of the GPT-4o Mini gives them another tool in their inventory. OpenAI allowed fintech startup Ramp to test the model, using the GPT-4o Mini to build a tool that extracts expense data from accounts. Instead of manually filling in text fields, users can upload an image of the receipt and the model sorts everything for them. Superhuman, an email client, also tested the GPT-4o Mini and used it to create an auto-suggest email reply feature.
The goal is to provide something light and cheap for developers to create all those applications and tools that they couldn’t afford with larger, more expensive models like the GPT-4. Many developers would opt for Claude 3 Haiku or Gemini 1.5 Flash before paying the huge costs of using the most robust models.
Why did OpenAI take so long? Godement said it was “pure prioritization” because the company was focused on creating bigger and better models like GPT-4, which required a lot of “human and computational resources.” As time went on, OpenAI noticed a trend of developers eager to use smaller models, so the company decided now is the right time to invest its resources in making the GPT-4o Mini.
“I think it will be very popular,” Godement said. “Both for existing applications that use all of OpenAI’s AI, and for many applications that were previously put off due to cost.”
The post OpenAI launches a cheaper and smarter model appeared first on ITNetwork.
Source: www.itnetwork.rs
