Microsoft Launches Phi 3.5 AI Models to Rival Google and OpenAI
Microsoft introduces three new Phi 3.5 AI models that outperform competitors on key benchmarks while remaining open source for commercial use.
Microsoft expands its AI capabilities beyond its OpenAI partnership by releasing three new models in its Phi series. The lineup includes the 3.82 billion parameter Phi-3.5-mini-instruct for fast reasoning, the 41.9 billion parameter Phi-3.5-MoE-instruct for complex tasks, and the 4.15 billion parameter Phi-3.5-vision-instruct for image and video analysis.
Developers access all three models on Hugging Face under a permissive MIT license that allows for unrestricted commercial usage and modification. Despite their relatively compact sizes compared to massive frontier models, these new releases achieve near state-of-the-art performance on third-party benchmark tests.
The Phi models even outperform industry heavyweights like Google's Gemini 1.5 Flash, Meta's Llama 3.1, and OpenAI's GPT-4o in specific evaluations such as long-context code understanding. The Phi-3.5-mini-instruct model shines in memory-constrained environments, while the Phi-3.5-MoE model introduces a new Mixture of Experts architecture to the series to handle more powerful reasoning tasks.