OpenAI released the lightweight large language model GPT-4o mini, which marks a new trend of miniaturization of large AI models and also reflects OpenAI's strategy of popularizing advanced AI technology. GPT-4o mini scored as high as 82% in the MMLU benchmark test, with excellent performance and extremely competitive pricing. It only costs 15 cents per million input tokens and 60 cents per million output tokens, which is an order of magnitude cheaper than previous top models. . OpenAI CEO Sam Altman believes that this marks that the cost of intelligence has become so low that there is no need to measure it.
As competition in the field of artificial intelligence becomes increasingly fierce, OpenAI once again leads the trend and launches a lightweight large language model called GPT-4o mini. This move not only marks a new trend of miniaturization of large AI models, but also is a concrete manifestation of OpenAI’s strategy to popularize advanced AI technology as widely as possible.
GPT-4o mini showed amazing performance upon its launch, scoring as high as 82% on the MMLU benchmark, surpassing its predecessor GPT-4. Even more impressive is its extremely competitive pricing strategy: only 15 cents per million input tokens and 60 cents per million output tokens, which is an entire order of magnitude cheaper than the previous top model. OpenAI CEO Sam Altman described this breakthrough as the cost of leading to intelligence is so low that there is no need to measure it.
The new model is not only available immediately on the free and paid versions of ChatGPT, but also supports text and visual input, and will be expanded to video and audio areas in the future. It has a context window of 128K tokens, knowledge as of October 2023. These features enable GPT-4o mini to outperform other small models in multiple benchmarks, including reasoning tasks, mathematical and coding capabilities, and multi-modal reasoning. .
In terms of security, OpenAI is not lax either. GPT-4o mini not only inherits the security measures of GPT-4o, but also undergoes rigorous expert evaluation and testing. It is worth mentioning that it is the first API model to apply an instruction layering approach, which significantly improves the model's ability to resist jailbreaking and prompt injection.
Market analysts generally believe that GPT-4o mini is likely to replace GPT-3.5 Turbo and directly compete with Claude3Haiku and Gemini1.5Flash. This not only reflects the increasingly fierce competition among AI technology suppliers, but also demonstrates OpenAI’s firm determination on the road to popularizing AI.
OpenAI's product manager Olivier Godement said the company's mission is to securely build and distribute AI to make it widely available. By continuously reducing costs while improving model capabilities, OpenAI is paving the way for a future where AI is seamlessly integrated into every application and website. With the launch of GPT-4o mini, developers now have a more economical and efficient choice, which is expected to promote the further popularization and innovation of AI applications.
Today, with the rapid development of AI technology, the advent of GPT-4o mini has undoubtedly injected new vitality into the entire industry. It not only represents technological progress, but also indicates that AI will be integrated into our daily lives in a more people-friendly manner. As costs continue to fall and performance continues to improve, we may be on the cusp of large-scale application of AI, witnessing the arrival of a smarter and more inclusive digital future.
The emergence of GPT-4o mini indicates that AI technology will serve the public at lower cost and higher efficiency, promote the popularization and innovation of AI applications, and lead us towards a more intelligent future.