Phancy Launches Phai-A1 Language Model and Phai-M1 Video Model
Phancy introduced the Phai-A1 language model and Phai-M1 video-generation model on November 18, targeting coding and assistant workflows as well as live-action video production. Phancy says Phai-A1 supports a native million-token context window and local PC deployment, while Phai-M1 can generate 15 seconds of 720P video with synchronized audio in about three minutes. The company also says both models underwent full post-training optimization on Huawei Ascend and will soon have their weights released on Hugging Face and ModelScope.
The launch combines local language-model inference, long-context support, and video generation with an end-to-end Huawei Ascend optimization path, but its headline performance claims and planned weight release remain unverified.

北京,2026年11月18日 -范式智能技术集团股份有限公司(“范式”或“集团”,股票代码:6682.HK,原第四范式)于APEC会议期间正式推出Phai系列两款自研模型:通用语言模型Phai-A1及视频生成模型Phai-M1,分别面向代码开发、智能助手等语言类场景,以及视频内容创作场景。
Phai-A1重点优化推理效率与长上下文处理能力,支持原生百万级上下文窗口,可在个人电脑本地部署;Phai-M1针对真人影视场景进行训练优化,重点改善人物呈现、画面细节和音画协同效果。两款模型分别从推理性能、视频生成品质出发,落地验证AI模型在真实业务流程中的应用价值。
范式创始人兼CEO戴文渊表示:“AI产业正从单一模型能力的竞争,走向算力、模型与应用协同发展的新阶段。推出Phai系列,是范式夯实AI全栈技术布局的重要一步。通过自研模型与国产算力平台的协同调优,我们能够打通底层算力与上层应用,让模型能力贴合业务需求,同时用真实场景的反馈持续反哺模型迭代,持续增强集团整体技术实力与产品竞争力。”
#Phancy#Phai-A1#Phai-M1#Qwen 27B#Huawei Ascend#Hugging Face#ModelScope#ModelHub XC#HAMi#ModelSphere#PhanthyCode#PhanthyMovie