OpenAI has released a new economical AI model GPT-4o mini. Its cost has been significantly reduced but its performance is not inferior. It marks a key step towards wider application of AI technology. This article will delve into the performance, security and price advantages of GPT-4o mini, as well as its impact on the future development of AI.
OpenAI has made another big move! Their latest GPT-4o mini is claimed to be the "most affordable" small model. This is not just a model upgrade, but the beginning of an intelligent revolution. Today, let us unveil the mystery of GPT-4o mini and see how it can make intelligence more "grounded".
Be smarter and save money
OpenAI’s vision is to make intelligence everywhere, and GPT-4o mini is the latest implementation of this vision. This model is not only significantly lower in cost, but also in terms of performance. At just 15 cents per million input tokens and 60 cents per million output tokens, it is an order of magnitude cheaper than previous cutting-edge models and more than 60% cheaper than GPT-3.5 Turbo.
Small stature, big wisdom
GPT-4o mini surpasses GPT-3.5 Turbo and other small models in academic benchmarks, both for text intelligence and multi-modal reasoning. It also supports the same language range as GPT-4o and excels in function calls, which enables developers to build applications that can obtain data or perform operations with external systems and improves compared to GPT-3.5 Turbo Improved long context performance.
On key benchmarks, GPT-4o mini performed as follows:
Reasoning tasks: In reasoning tasks involving text and vision, GPT-4o mini scored 82.0%, compared to 77.9% for Gemini Flash and 73.8% for Claude Haiku.
Mathematics and Coding Ability: GPT-4o mini also performed well in mathematical reasoning and coding tasks. In the MGSM (mathematical reasoning) test, it scored 87.0%, compared to 75.5% for Gemini Flash and 71.7% for Claude Haiku. In the HumanEval (encoding performance) test, it scored 87.2%, compared with 71.5% for Gemini Flash and 75.9% for Claude Haiku.
Multimodal Reasoning: In MMMU (Multimodal Reasoning Evaluation), GPT-4o mini scored 59.4%, while Gemini Flash scored 56.1% and Claude Haiku scored 50.2%.
Built-in security measures
Security is always at the core of openAI model development. During the pre-training phase, openAI filters out information that it does not want the model to learn or output, such as hate speech, adult content, websites that mainly aggregate personal information, and spam. After training, openAI uses techniques such as reinforcement learning and human feedback (RLHF) to align the model's behavior with openAI's policies and improve the accuracy and reliability of the model's response.
GPT-4o mini has the same security mitigations built into GPT-4o, which openAI carefully evaluated through automated and human evaluation based on the original readiness framework and voluntary commitments. More than 70 external experts in areas such as social psychology and misinformation tested GPT-4o to identify potential risks, which openAI has now addressed and plans to include in the upcoming GPT-4o System Card and Readiness Score Card. Share details. Insights from these expert assessments have helped improve the security of GPT-4o and GPT-4o mini.
Availability and pricing
GPT-4o mini is now available in the Assistant API, Chat Completion API, and Batch API as text and visual models. Developers pay 15 cents per 1M input tokens and 60 cents per 1M output tokens (roughly equivalent to 2500 pages in a standard book). We plan to roll out fine-tuning capabilities for GPT-4o mini in the coming days.
In ChatGPT, Free, Plus and Team users will be able to access GPT-4o mini starting today, replacing GPT-3.5. Enterprise users will also have access starting next week, in line with openAI’s mission to make the benefits of AI available to everyone.
future outlook
The OpenAI team said: “Over the past few years, we have witnessed significant advances in AI intelligence while dramatically reducing costs. For example, since the launch of the less powerful text-davinci-003 model in 2022, GPT-4o mini’s Cost per token has dropped by 99%. We are committed to continuing to reduce costs while enhancing model capabilities."
“We envision a future where models are seamlessly integrated into every app and every website. GPT-4o mini paves the way for developers to build and scale powerful AI applications more efficiently and affordably. The future of AI is becoming more accessible, reliable, and embedded in our daily digital experiences, and we’re excited to continue to lead the charge.”
All in all, GPT-4o mini provides a solid foundation for the popularization of AI applications with its excellent performance, economical price and strong security measures, indicating that AI technology will be more widely integrated into our lives.