The GPT-4o mini, OpenAI’s most reasonably priced small-scale language model, is available for purchase at 15 cents per million input tokens and 60 cents per million output tokens. The model aims to make sophisticated AI more accessible to developers and companies by outperforming competitors on benchmarks and including better safety features.
GPT-4o mini, OpenAI’s most affordable small-scale language model yet, has been released.
The GPT-4o mini, which is priced at under 15 cents per million input tokens and 60 cents per million output tokens, is more than 60% less expensive than the GPT-3.5 Turbo model and is intended to greatly broaden the spectrum of AI-powered applications.
“Our mission is to make intelligence as broadly accessible as possible. GPT-4o Mini is a major step forward in that direction, offering developers and businesses a cost-effective way to leverage the power of large language models,” said Sam Altman, CEO of OpenAI.
In a blog post, OpenAI claimed that the GPT-4o mini outperforms other small-scale models available on the market, although it is less expensive. The new model outperforms Gemini Flash (77.9%) and Claude Haiku (73.8%) with an 82% score on the Multimodal Language Understanding (MMLU) test, according to OpenAI’s internal benchmarks.
“GPT-4o Mini is a game-changer because of its cost-effectiveness and robust performance on various tasks. We expect this model to enable a whole new wave of AI-powered applications and services,” said Dario Amodei, Head of Research at OpenAI.
“The model’s ability to reason across text, vision, and other modalities opens up a wealth of new use cases,” Altman stated. “Developers can now build applications that seamlessly integrate AI-powered analysis of images, videos, and audio alongside natural language processing.”
Enhanced reliability and safety
Using the knowledge it gained from deploying its larger GPT-4o model, OpenAI has incorporated numerous safety features into the GPT-4o mini.
The new model uses reinforcement learning techniques to align the model’s behavior with OpenAI’s safety regulations and improved filtering to remove potentially dangerous information during pre-training. Furthermore, the GPT-4o mini is the first model to use the instruction hierarchy mechanism developed by the business, which enhances the device’s resilience against quick injections and jailbreaks.
“We envision a future where AI-powered intelligence is seamlessly integrated into every app and website. GPT-4o Mini is a significant step towards that vision, making it more accessible for developers to build innovative, AI-driven solutions,” Altman declared.
The model may now be accessed via OpenAI’s Assistants API, Chat Completions API, and Batch API. In the next few days, fine-tuning features will be made accessible. Starting today, ChatGPT users will also have access to the GPT-4o mini, which is a replacement for the GPT-3.5 Turbo model.
