OpenAI Unveils Cost-Effective ‘Mini’ Model, GPT-4o Mini, to Enhance AI Accessibility Amid Rising Competition

To go back to this article, navigate to My Profile, and then look at your saved stories.

OpenAI Introduces Budget-Friendly ‘Mini’ Model for Broader AI Access

Today, OpenAI unveiled a more affordable “mini” version of its model, aimed at enabling wider access to its artificial intelligence technology for businesses and applications. The newly launched GPT-4o mini, which is accessible from today, is priced 60 percent lower than the least expensive model previously available from OpenAI, yet delivers improved performance, according to the company.

OpenAI describes this initiative as a step towards ensuring AI technology is widely available to everyone. This decision also highlights the increasing rivalry between AI cloud service providers and the surging demand for small, free, open-source AI frameworks. Meta is anticipated to launch its most advanced and complimentary version, Llama 3, in the upcoming week.

"OpenAI's primary mission is to create and share artificial intelligence in a secure manner while ensuring it's widely available," explains Olivier Godement, who oversees the new model at OpenAI, to WIRED. "One of the key strategies for achieving this is by offering intelligence solutions at a reduced price."

Godement reports that the company has created a more affordable product by enhancing the design structure and optimizing the data used for training, as well as the training process itself. According to OpenAI, GPT-4o mini surpasses competing "small" models in various standard tests.

OpenAI has successfully established a strong presence in the cloud-based artificial intelligence sector, largely due to the impressive features of its chatbot, ChatGPT, introduced towards the end of 2022. The organization provides external users with the opportunity to utilize the advanced language model behind ChatGPT, known as GPT-4o, in exchange for a fee. Additionally, it makes available a more affordable, albeit less advanced, version known as GPT-3.5 Turbo, at approximately one-tenth the price of GPT-4o.

The surge in fascination with language models, sparked by the phenomenal popularity of ChatGPT, has led rival companies to create comparable products. Google, a trailblazer in the field of AI, has aggressively moved forward with its efforts to create and bring to market a significant language model and chatbot, known as Gemini. Meanwhile, startups like Anthropic, Cohere, and AI21 have secured millions in funding to craft and promote their large language models targeted at businesses and developers.

Developing top-tier large language models involves significant investment, yet certain firms have opted to make their developments public. This strategy aims to draw developers towards their platforms. Among these open-source AI models, Meta's Llama stands out. It is available for free download and utilization, although its license includes restrictions on its use for commercial purposes.

In April, Meta unveiled Llama 3, its latest and most advanced open-source AI model. The tech giant made available a compact version of this model, featuring 8 billion parameters, which serves as an indicator of the model's sophistication and adaptability. Additionally, a more robust, intermediate version was launched, boasting 70 billion parameters. This version competes closely with OpenAI's top model in terms of performance on numerous standard evaluations.

Multiple informants have disclosed to WIRED that Meta is set to unveil its most extensive iteration of Llama 3, featuring 400 billion parameters, on July 23, though they note the launch timing may shift. The potential capabilities of this specific version of Llama 3 remain uncertain, yet various firms are increasingly drawn to open-source AI models due to their affordability, adaptability, and the enhanced command over both the model itself and the input data it processes.

Godement acknowledges the changing demands of consumers. He explains, "There's a growing trend among developers and companies to blend both compact and expansive models to craft the optimal product experience, balancing cost and response time to their advantage," he notes.

Godement indicates that the cloud services offered by OpenAI feature models that undergo more extensive security assessments compared to those of its rivals. He also mentions the possibility of OpenAI creating models that could be operated directly on users' devices in the future. "Should there be a significant interest, we might consider making that option available," he states.

Explore More…

Dive into the electoral season with our WIRED Politics Lab newsletter and podcast.

Unsure if breakdancing qualifies as an Olympic event? The global champion shares a similar sentiment (sort of).

Investigators decrypted a decade-old passphrase unlocking a cryptocurrency wallet valued at $3 million

The surprising emergence of the first-ever beauty contest judged by artificial intelligence

Ease the strain on your spine: Discover the top-rated office chairs from our evaluations.

Additional Content from WIRED

Insights and Tutorials

© 2024 Condé Nast. All rights reserved. WIRED has the potential to receive a share of revenue from products bought via our website, a result of our Affiliate Partnerships with store owners. The content on this website is protected and cannot be copied, shared, transmitted, stored, or utilized in any form without explicit consent from Condé Nast. Advertisement Choices

Choose a global website