Microsoft has released three new AI models in its Phi series: Phi-3.5-mini-instruct (3.82 billion parameters), Phi-3.5-MoE-instruct (41.9 billion parameters), and Phi-3.5-vision-instruct (4.15 billion parameters).

The Phi-3.5-mini-instruct supports compute-constrained environments, supporting a 128,000 token context length for code generation and mathematical problem-solving. The Phi-3.5-MoE-instruct uses a "Mixture of Experts" architecture for reasoning tasks across multiple languages. The Phi-3.5-vision-instruct integrates text and image processing capabilities for image understanding, optical character recognition, and video summarization.

Analyst QuickTake: Microsoft introduced the Phi-3-vision, a small language model capable of analyzing images and text, in May 2024, and Phi-3 Mini, a lightweight AI model that can understand complex instructions like larger models, in April 2024. These latest AI release underscores the company’s commitment to AI innovation targeting diverse application needs.