OpenAI is releasing a new, more affordable AI model called GPT-4o Mini. This lighter version costs significantly less than the full-sized models and is more capable than GPT-3.5. The high costs associated with building apps using OpenAI’s models often deter developers without substantial budgets. To address this, OpenAI is entering the light model market, challenging cheaper alternatives like Google’s Gemini 1.5 Flash and Anthropic’s Claude 3 Haiku.
Olivier Godement, who leads the API platform product, emphasized that GPT-4o Mini aligns with OpenAI’s mission of making AI more accessible. “If we want AI to benefit every corner of the world, every industry, every application, we have to make AI much more affordable”.
GPT-4o Mini is now available to ChatGPT users on Free, Plus, and Team plans, replacing GPT-3.5 Turbo. Enterprise users will gain access next week. While GPT-3.5 will still be available via the API, it will eventually be retired. Godement believes the new model will be very popular among developers.
The lightweight model supports text and vision in the API, with future capabilities to handle all multimodal inputs and outputs like video and audio. Despite these features, GPT-4o Mini is designed for simpler tasks, not sophisticated virtual assistants.
In terms of performance, GPT-4o Mini achieved an 82 percent score on the Measuring Massive Multitask Language Understanding (MMLU) benchmark, which includes around 16,000 multiple-choice questions across 57 academic subjects. This score is a significant improvement over GPT-3.5’s 70 percent. For comparison, GPT-4o scored 88.7 percent, while Google’s Gemini Ultra holds the highest-ever score of 90 percent. Competing models Claude 3 Haiku and Gemini 1.5 Flash scored 75.2 percent and 78.9 percent, respectively.
Despite these benchmarks, researchers remain cautious about their reliability due to varying administration methods and potential dataset overlaps. This skepticism highlights the challenges in making accurate comparisons between different models.
Developers eager to create affordable AI applications now have a new tool with GPT-4o Mini. OpenAI allowed the financial technology startup Ramp to test the model, using it to build a tool that extracts expense data from receipts. Similarly, the email client Superhuman used GPT-4o Mini to create an auto-suggestion feature for email responses.
The goal of GPT-4o Mini is to provide a lightweight, cost-effective option for developers to build apps and tools that were previously too expensive to develop with larger models like GPT-4. Many developers have previously turned to Claude 3 Haiku or Gemini 1.5 Flash to avoid high compute costs.
Godement explained that the delay in releasing a lighter model was due to “pure prioritization,” as the company focused on developing bigger and better models like GPT-4. Recognizing the growing demand for smaller models, OpenAI decided it was time to invest in GPT-4o Mini.
“I think it’s going to be very popular,” Godement reiterated, “both by existing apps that use all the AI at OpenAI and also many apps that were priced out before.”
With GPT-4o Mini, OpenAI is making strides in democratizing AI, providing a powerful yet affordable tool for developers worldwide.
