Phi-4-instruct (reasoning-trained)
Phi-4-instruct (reasoning-trained) is an AI model published by Microsoft Azure, released on 2025-02-27, for 推理大模型, with 38.0B parameters, and 128K tokens context length, requiring about 7.67GB storage, under the 不开源 license.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
Phi-4-instruct (reasoning-trained) currently shows benchmark results led by AIME 2024 (47 / 62, score 50), MATH-500 (36 / 43, score 90.40), GPQA Diamond (143 / 166, score 49). This page also consolidates core specs, context limits, and API pricing so you can evaluate the model from benchmark results and deployment constraints together.
微软近期发布的 Phi-4-mini-instruct 模型是一款轻量级的开放性语言模型，属于 Phi-4 系列，并专注于多语言支持与高效推理能力。它在基于合成数据和公开网站数据进行训练时，特别注重高质量推理密集型数据。这款模型不仅具有较强的推理能力，还在多语言任务中表现出色，支持高达 128K tokens 的上下文长度。
Phi-4-multimodal-instruct (reasoning-trained)是经过思维链训练过的，其推理能力更强，其基座版本参考： https://www.datalearner.com/ai-models/pretrained-models/Phi-4-mini-instruct
不过，该模型仅仅在微软的技术报告中提及，并未开源发布。
欢迎关注 DataLearner 官方微信，获得最新 AI 技术推送
