2026年夏季，人工智能在数学领域实现了突破性进展。OpenAI的Astra模型在8月4日公开了62页手稿，展示其对十大菲尔兹奖级别数学难题的突破，完成这些推演的成本仅为2000美元[1]。与此同时，月之暗面于7月17日发布了全球参数规模最大的开源模型Kimi K3，该模型拥有2.8万亿参数，并在7月27日实现全量开源，支持100万字上下文处理能力[1]。
在AI能力突飞猛进的背景下，科技业界对开源模型的政策导向也发生了变化。英伟达创始人黄仁勋于7月24日在X平台表示，"这个世界既需要最前沿的闭源模型，也需要最前沿的开放模型，我们不能制裁开源"[1]，联合多家科技公司反对对开源模型施加监管限制，强调开源对AI生态的重要性。这一立场与OpenAI过去的态度形成对比——2019年OpenAI曾因担心被滥用而拒绝开源GPT-2完整模型[1]。
OpenAI's latest breakthrough has intensified debates about artificial intelligence's expanding capabilities in mathematics. The company's Astra model publicly demonstrated on August 4th a manuscript spanning 62 pages that showcases solutions to ten Fields Medal-level mathematical problems, with the cost of mathematical derivation totaling merely 2,000 US dollars [1]. This achievement underscores the rapid advancement of AI in domains traditionally reserved for elite human mathematicians.
The same period witnessed parallel developments in the open-source AI landscape. On July 17th, Moonshot AI released Kimi K3, which achieved full open-source status on July 27th [1]. The model, featuring 2.8 trillion parameters, became the world's largest open-source language model and supports processing up to one million characters of context [1]. These technical milestones have reignited discussions about the direction of AI development and regulatory approaches.
Amid these developments, influential voices in the technology sector have voiced opposition to potential regulatory constraints. Nvidia founder Jensen Huang posted on the X platform on July 24th, stating: "This world needs both the most cutting-edge closed-source models and the most cutting-edge open models. We cannot sanction open source" [1]. Huang's position reflects growing consensus among industry leaders that open-source development remains essential to the AI ecosystem. This stance contrasts with historical precedent: in 2019, OpenAI had refused to open-source the complete GPT-2 model due to concerns about potential misuse [1].