Top 3 Phi 3.5 Models by Microsoft: Small But Powerful AI Innovations

Explore Microsoft's new Phi 3.5 models—Mini Instruct, MoE, Vision Instruct—available on Hugging Face for advanced AI tasks with an MIT license.

Today, Microsoft officially upgraded its "small language model" series, releasing three new models under Phi 3.5:

• Phi-3.5-mini-instruct with 3.82B parameters

• Phi-3.5-MoE-instruct with 41.9B parameters

• Phi-3.5-vision-instruct with 4.15B parameters

Microsoft's collaboration with OpenAI has been successful, but they are clearly not satisfied. They continue to push the boundaries in large models, consistently innovating and advancing the field.

These models are available for download, use, and fine-tuning on Hugging Face, with a permissive MIT license from Microsoft, allowing unrestricted commercial use and modification.

Despite their relatively small size, these models perform well in many third-party benchmarks, even outperforming other leading large models like Google’s Gemini 1.5 Flash, Meta’s Llama 3.1, and in some cases, OpenAI’s GPT-4o.

This performance advantage, combined with the open-source license, has garnered significant praise for Microsoft on social network X.