- Microsoft announces new AI model Phi-4
- It is now available to developers and researchers
- Performs well on mathematical tasks despite the small scale
Microsoft has announced a brand new AI model called Phi-4, a small language model (SLM) as opposed to the large language models (LLM) that chatbots like ChatGPT and Copilot use. Phi-4 is not only lightweight, but also excels at complex reasoning, making it perfect for math and language processing.
Microsoft has released a series of benchmarks showing that Phi-4 outperforms even large language models like Gemini Pro 1.5 on math competition problems.
Post-training breakthroughs
Small language models, such as ChatGPT-4o mini, Gemini 2.0 Flash, and Claude 3.5 Haiku, are typically faster and cheaper to use compared to large language models. However, their performance has increased dramatically with recent versions.
For Microsoft, these improvements may have come about through breakthroughs in training Phi-4 on high-quality synthetic datasets and post-training innovations. Because the bottleneck to improving AI capability has always been the sheer amount of processing power and data required for training (also called the “pre-training data wall”), AI companies have instead looked at ways to improve development after training. to improve performance.
Phi-4 is currently available on Azure AI Foundrya platform for developers to build generative AI applications. So while Phi-4 is available under a Microsoft Research license agreement, you can’t just start chatting with it like you would with Copilot or ChatGPT. Instead, we’ll have to wait and see what people produce with it in the future.