消息称 DeepSeek V4 模型打破惯例：华为等国内厂商可早期访问，不让英伟达 AMD 先用

Group 1 - DeepSeek has not shared its upcoming flagship model with US chip manufacturers like Nvidia and AMD, which breaks industry norms [1][4][5] - Instead, DeepSeek V4 has provided early access to domestic suppliers, including Huawei Technologies [1][4] - AI developers typically share pre-release versions of major models with chip manufacturers to ensure software efficiency on widely used hardware [5] Group 2 - DeepSeek is currently testing the V4 Lite model, codenamed "Sealion-lite," which features a context window of 1 million tokens and supports native multimodal reasoning [5] - The latest update from DeepSeek has expanded its knowledge base to May 2025, allowing it to accurately output news from April 2025 in offline mode [2][5]