In recent years, the cost of large language models (LLMs) has been a major factor hindering their widespread application. Recently, Mistral and Microsoft have led the trend of "small language models", which has brought new hope for reducing the cost of AI applications and expanding the application scope of generative AI technology. The Mistral-medium model even surpasses GPT-4 in terms of code capabilities, and the cost is reduced by two-thirds, which is undoubtedly a milestone progress. The release of the Mixtral8x7B model and the Phi-2 small model has further promoted this trend.
Mistral and Microsoft are leading the trend of "small language models". Mistral-medium's code capabilities surpass GPT-4 and its cost is reduced by 2/3. The release of Mixtral8x7B model and Phi-2 small model has made the trend of small models more and more popular, reducing the cost of large-scale AI applications and broadening the application scope of generative AI technology. Mistral-medium internal testing results show that its code generation quality and cost are better than GPT-4, providing developers with a more efficient choice.The rise of small language models provides more economical and efficient solutions for the popularization and application of AI technology, and also heralds a new direction for the future development of AI. The successful case of Mistral-medium will undoubtedly encourage more companies and research institutions to invest in the research and development and application of small models, thereby promoting artificial intelligence technology to benefit human society faster and better. In the future, we have reason to look forward to the emergence of more and better small language models, which will have a transformative impact on all walks of life.