【文/观察者网专栏作者 心智观察所】 几天前，《Nature》杂志刊发了一篇来自中国的人工智能研究论文。这在顶级学术期刊上并非新鲜事， 但这篇论文的分量却非同寻常：它来自北京智源人工智能研究院，核心成果是一个名为"Emu3"的多模 态大模型，而它试图回答的问题，是整个AI领域过去五年来悬而未决的核心命题——我们能否用一种 统一的方式，让机器同时学会看、听、说、写，乃至行动？ 这个问题听起来简单，但它的复杂程度足以让全球顶尖的AI实验室争论不休。 这个选择的意义，可能需要一些背景知识才能理解。 | https://doi.org/10.1038/s41586-025-10041-x | Xinlong Wang148, Yufeng Cui14, Jinsheng Wang14, Fan Zhang14, Yu | | --- | --- | | Received: 11 November 2024 | Xiaosong Zhang14, Zhengxiong Luo14, Quan Sun14, Zhen Li14, Yuqi | | | Yingli Zhao', Yulong Ao', Xuebin Mi ...

Core Insights - The recent launches of AI models by Google and Anthropic signify a competitive shift in the AI landscape, with both companies aiming to enhance their market positions through innovative features and cost reductions [1][3][5] Company Developments - Google launched Gemini 3 on November 18, emphasizing advancements in multimodal reasoning and visual understanding, aiming to regain leadership in the AI sector [1] - Anthropic introduced Claude Opus 4.5 six days later, claiming it outperformed human candidates in internal assessments, showcasing its capabilities in coding and long-horizon reasoning [3][7] Cost Efficiency - Both companies have significantly reduced operational costs for their new models, with Anthropic cutting the price of Claude Opus 4.5 by 67%, from $15 to $5 per million tokens, while Google set Gemini 3 Pro at $2 for reading and $12 for generation [4][5] Model Capabilities - Gemini 3 excels in processing various data types, achieving over 90% on the GPQA Diamond benchmark for scientific reasoning, which could transform workflows involving design and video feedback [6] - Claude Opus 4.5 focuses on coding and complex data analysis, outperforming Gemini 3 Pro in real engineering tasks and demonstrating strong consistency in extended sequences [7][10] Market Positioning - The pricing strategies of both models reflect a rapid shift in the economics of high-end AI, allowing for broader usage across workflows [5] - Gemini 3 is integrated into Google's broader ecosystem, enhancing its capabilities in search and development platforms, while Claude Opus 4.5 is paired with new product integrations for tools like Excel [9][8] Production-Level Execution - Both models are designed for multistep tasks rather than isolated responses, with Gemini 3 demonstrating superior decision-making in a business simulation benchmark [11][12]