Microsoft launches Phi-4 AI model that can beat Google Gemini Pro 1.5 in this ‘key’ area

Microsoft has unveiled an AI model, Phi-4, a compact yet powerful language model that is claimed to have achieved remarkable mathematical reasoning capabilities while using significantly fewer computational resources than its larger competitors.

At a time when companies have been locked in a race to build ever-larger models with billions or even trillions of parameters, Phi-4, with its streamlined 14-billion-parameter architecture, demonstrates that smaller, well-designed models can deliver superior performance in specific domains, such as mathematical reasoning.

“Phi-4 outperforms comparable and larger models on math related reasoning due to advancements throughout the processes, including the use of high-quality synthetic datasets, curation of high-quality organic data, and post-training innovations. Phi-4 continues to push the frontier of size vs quality,” Microsoft said.

The company also claims that Phi-4 outperforms much larger models, including Gemini Pro 1.5, on math competition problems.

Microsoft’s Phi-4 is efficient

Large language models require significant computing power, driving up costs and energy consumption. Phi-4’s smaller size makes it more accessible to organisations with limited budgets and computing resources.

Microsoft is taking a cautious approach to Phi-4's release, initially making it available through its Azure AI Foundry platform under a research license agreement.

“Phi-4 is currently available on Azure AI Foundry under a Microsoft Research License Agreement (MSRLA) and will be available on Hugging Face next week,” Microsoft added.

This controlled rollout includes safety features and monitoring tools to address concerns about AI risks and misuse.

"Building AI solutions responsibly is at the core of AI development at Microsoft. We have made our robust responsible AI capabilities available to customers building with Phi models, including Phi-3.5-mini optimized for Windows Copilot+ PCs," it added.