Early this morning, Microsoft Research open-sourced the strongest small parameter model to date—phi-4. On December 12 last year, Microsoft first showcased phi-4, which has only 14 billion parameters but performs exceptionally well. It outperformed OpenAI's GPT-4o in GPQA graduate-level assessments and the MATH benchmark tests, as well as top open-source models Qwen 2.5 -14B and Llama-3.3-70B. In the USA Mathematics Competition AMC test, phi-4 achieved a score of 91.8, surpassing well-known open and closed source models like Gemini Pro 1.5, GPT-4o, Claude 3.5 Sonnet, and Qwen 2.5, with overall performance comparable to the 405 billion parameter Llama-3.1. (AIGC open Community)