Microsoft continues its impressive success streak in the field of artificial intelligence and has announced the release of three new models in the Phi-3.5 series today. These models, which are characterized by advanced multimodality and multilingual capabilities, aim to further revolutionize the market for AI-based applications. The models have been made available on Hugging Face under a Microsoft-branded MIT license, providing developers worldwide the opportunity to freely use, adapt, and further develop these innovative technologies.

The three models – Phi-3.5-mini-instruct, Phi-3.5-MoE-instruct, and Phi-3.5-vision-instruct – cover a wide range of applications, from basic to highly complex tasks. Each model is optimized for specific requirements, such as fast and precise reasoning or processing text and image data in multimodality tasks.

The Phi-3.5 Mini Instruct Model, equipped with 3.8 billion parameters, is a lightweight model specifically designed for use in memory or computation-constrained environments. It demonstrates impressive performance on tasks requiring strong reasoning, such as code generation, mathematical problem-solving, and logic-based queries. Despite its compact size, it outperforms other models in its class, like the Llama-3.1-8B-instruct, on the RepoQA benchmark, especially on tasks requiring an understanding of long contexts.

The Phi-3.5 MoE (Mixture of Experts) Model is the first of its kind in Microsoft's portfolio.

The Phi-3.5 Vision Instruct Model integrates text and image processing capabilities, making it ideal for tasks such as general image processing, optical character recognition, and video summarization. With support for 128k token context lengths, this model can handle complex, multi-layered visual tasks. Microsoft emphasizes that the model was trained on a combination of synthetic and publicly available datasets, with a focus on high-quality and reasoning-intensive data.

All three models of the Phi-3.5 series have been released under an MIT license, underscoring Microsoft's commitment to supporting the open-source community. This license allows developers to freely use, modify, and distribute the software, while also adhering to the disclaimers of Microsoft and other copyright holders.

The Release of the Phi-3.5 Models Represents a Significant Advancement in the Development of Multilingual and Multimodal AI. With these models, Microsoft provides developers the opportunity to integrate state-of-the-art AI capabilities into their applications, promoting innovations in both commercial and research fields.