Core Viewpoint - The article discusses the significant upgrades of Google's Gemini 3.1 Pro model compared to its predecessor, Gemini 3 Pro, highlighting improvements in multimodal generation, semantic understanding, and reasoning capabilities [1][9][10]. Group 1: Model Upgrades - Gemini 3.1 Pro shows a noticeable enhancement in multimodal generation and semantic understanding, achieving a higher level of performance [1]. - The model can convert everyday data into interactive visual content, such as aerospace dashboards and city simulations [3][5]. - In the ARC-AGI-2 benchmark test, Gemini 3.1 Pro achieved a verification score of 77.1%, which is double that of Gemini 3 Pro [10]. Group 2: Performance Metrics - The performance comparison table indicates that Gemini 3.1 Pro outperforms other models in various benchmarks, including academic reasoning and abstract reasoning puzzles [11]. - The overall ranking score of Gemini 3.1 Pro in Arena's evaluation is 13 points higher than that of Gemini 3 Pro, with significant improvements in text and code dimensions [12]. - The model supports a context length of 1 million tokens and has a knowledge cutoff date of January 2025, enhancing its multimodal understanding and long-context performance [11]. Group 3: User Experience and Applications - Users have reported positive experiences with Gemini 3.1 Pro, generating complex visualizations and interactive applications, such as a 3D simulation of a flock of birds [17][20]. - The model has been utilized to create personal websites and educational applications, showcasing its versatility and advanced capabilities [24][25]. - The model is now available in Gemini applications and APIs, with specific access for Google AI Pro and Ultra users [29]. Group 4: Cost and Market Implications - The release of Gemini 3.1 Pro marks Google's first use of a ".1" version number, indicating a rapid pace of development in large models [30]. - The pricing for Gemini 3.1 Pro remains competitive, with input costs at $2 for less than 200k tokens and $4 for more, while output costs are $4 for less than 200k tokens and $18 for more [36]. - The cost per ARC-AGI-2 task is approximately $0.96, significantly lower than the previous model, suggesting a shift in the cost-performance curve in AI development [37][41].

编辑部 发自 凹非寺 量子位 | 公众号 QbitAI AI产业方向 岗位职责： AI产业方向 ：关注基建层创新，包含芯片、AI Infra、云计算； AI财经方向 ：关注AI领域创投和财报，跟踪产业链资本动向； AI产品方向 ：关注AI在应用和硬件终端方向的进展。 AI热潮还在汹涌，但如果你还不知道如何参与……那为什么不来 量子位 呢？ 我们是一家以 追踪AI新进展 为核心的内容平台，经过8年积累，目前拥有顶流影响力，广泛且备受认可的产业资源，以及时代风口的最佳观 测和学习生态位。 目前，我们有 三大方向 岗位招聘，希望你是 （或者能成为） 这三个方向的内容专家： 岗位均为全职，工作地点：北京中关村。 岗位面向： 加入我们，你可以获得： 以下是岗位详情： 所有岗位不同能力层级职位均在开放，欢迎结合个人履历和经验申请。 社招：覆盖编辑、主笔、主编各个层级，按能力匹配岗位； 校招：应届毕业生，接受实习且可转正。 站在AI浪潮之巅 ：第一时间接触和了解AI领域最新技术和产品，构建完整的AI认知体系。 玩转AI新工具 ：将各种AI新技术、新工具应用于工作，提升工作效率和创造力。 打造个人影响力 ：通过撰写独家原创内 ...

一水 发自 凹非寺 量子位 | 公众号 QbitAI 10亿融资，50亿估值。 这就是李飞飞创业公司 World Labs 最新交出的答卷。 今年1月，这家炙手可热的世界模型创业新星就被传出可能正在进行一轮5亿美元新融资，但最终结果却远超预期—— 不是5亿，而是10亿美金 （约合人民币69亿元） ，而且这一轮AMD、英伟达、富达等都投了。 加上本轮融资，成立于2024年4月的World Labs可谓成长飞速—— 不到两年时间，估值从最初的10亿美金直接翻了5倍 。 就这么说吧，可能最接近World Labs融资速度的Anthropic，也花了25个月才达到50亿美元估值。 这速度，你品，你细品。 不到两年，估值一路飙升 进入2026，谁都知道世界模型很hot，但看到World Labs成长如此之快，人们或许才对这一点有了更多实感。 短短一年多时间，这家公司经历了5倍重估。 虽然这里面不乏AI教母李飞飞带来的光环效应，但不得不说，投资者对世界模型的押注决心，比预想中来得更早、也更猛。 这一点，从World Labs本轮投资阵容也可一窥。 根据官方公告，本轮融资吸引了AMD、Autodesk、Emerson Co ...

克雷西 发自 凹非寺 量子位 | 公众号 QbitAI 刚刚，Gemini摇身一变，成了专业的"唱作人"。 谷歌把最新的Lyria 3模型塞进了Gemini，直接在对话框里招呼一声，Gemini就能现场给你攒个乐团。 这套玩法主打一个有手就行，给它打一段天马行空的文字，或者干脆甩过去一张刚拍的照片，它几秒钟内就能根据你的想法，吐出一首带歌 词、旋律甚至人声演唱的完整作品，整个过程快得惊人。 而且还顺便拉来了Nano Banana模型当帮手，曲子刚生成完，一张风格特搭的专辑封面也就跟着出炉了。 总之，从你想出点子到拿到带封面的专属BGM，中间的步骤简化到了极点。 网友评价，48kHz的立体声质量，加上根据照片生成音乐的功能，可见DeepMind这一波非常注重创意工作流程。 你的照片能开口唱歌了 硬指标上，Lyria 3的音频采样率来到了48KHz的高保真级别。这种规格让生成的曲子底气特别足，每一声琴鸣都显得非常扎实，有了这个音 质底座，看图唱曲的功能才更有发挥空间。 你随手上传一张在森林徒步的照片，AI就能瞬间捕捉到那种静谧感，转手给你配上一段对味的民谣，让原本静止的风景瞬间有了自己的声音。 这下，你的朋友圈也 ...

Core Viewpoint - The article discusses the integration of AI technology and embodied intelligence into mainstream culture during the 2026 New Year's Eve, highlighting the need for technology companies to maintain engagement beyond initial exposure during the Spring Festival Gala [1][4]. Group 1: AI and Embodied Intelligence Engagement - The Spring Festival Gala served as a peak moment for AI and robotics, but companies are anxious about sustaining interest beyond the event [4][5]. - Following the gala, tech companies are actively seeking to extend discussions and maintain engagement through online platforms, particularly Bilibili [6][7]. - Bilibili is emerging as a key platform for AI and robotics discussions, capitalizing on the momentum generated during the Spring Festival [8][30]. Group 2: Bilibili's Role in AI and Robotics - Bilibili's collaboration with the Spring Festival Gala has deepened, with the platform becoming the exclusive bullet screen video platform for the event [33]. - The platform has a unique ecosystem that fosters discussions around AI and robotics, making it a prime location for brands to engage with a tech-savvy audience [37][39]. - Data indicates that Bilibili has a high concentration of users interested in AI, with nearly 100,000 active creators related to AI content each month [38][39]. Group 3: User Engagement and Content Creation - The introduction of "interest rooms" on Bilibili allows for more targeted engagement, facilitating a smoother transition from awareness to deeper understanding of AI and robotics [62][63]. - Users are encouraged to interact with content, leading to a more engaged community that discusses, creates, and shares AI-related content [60][64]. - The active bullet screen culture on Bilibili enhances the information density of tech discussions, making it a unique platform for real-time feedback and interaction [52]. Group 4: Long-term Strategy for Brands - Companies like Songyan Power, Yuanbao, and Yushu Technology are not just marketing for the Spring Festival but are strategically positioning themselves for long-term brand recognition and user engagement [69][70]. - The article emphasizes the importance of finding the right community to sustain interest and deepen user understanding of AI and robotics post-exposure [70][71]. - The future of human-machine coexistence will depend on who can effectively engage the most receptive and innovative audience [68].

克雷西 发自 凹非寺 量子位 | 公众号 QbitAI 现在想找个既能干活又像真人一样好聊的模型变难了，AI好像正在变得越来越理性，但也越来越"不通人性"。 在这个节骨眼上，蚂蚁百灵大模型家族全新推出了万亿参数的旗舰级模型 Ling-2.5-1T ，不仅主打通用全能，还是个能够高效回复的即时模 型。 具体来说，Ling-2.5-1T 既拥有强大的Agent执行力，又保留了情商和写作能力 。 同时，它还想证明万亿参数的大块头也能身轻如燕，不需要在那儿"转圈思考"半天才能出结果，关键还不喜欢废话，非常节约Token。 在与前一代、及现在主流的大尺寸即时模型对比中，Ling-2.5-1T 在复杂推理、 指令遵循能力方面具有明显优势 。 | | Benchmark | Evaluation Config | Ling-2.5-1T | Ling-1T | 国产模型A | 国产模型B | GPT-5.2-chat | | --- | --- | --- | --- | --- | --- | --- | --- | | | | | | | (非思考) | (非思考) | | | | C-SimpleQA | Acc | ...

梦瑶 发自 凹非寺 量子位 | 公众号 QbitAI 今年春晚到底有多卷？卷到——机器人都开始抢镜头了。（震惊.jpg） 在今年春晚的 宜宾分会场 上，一群超聪明又带派的小机器人们，直接把现场氛围拉满～ 先看这个 托马斯360度回旋 特技，上一秒低位扫场，下一秒高位稳住，主打一个：我贼能转，你别眨眼～ 再来看这个， 全球首次 百台机器熊猫协同奔跑跳舞，超拟真不说，步伐整齐到还真有点让人上头！ 不光会特技，干活更见真章，瞧瞧这个 捞面大师 ，起面、控水、倒面一气呵成，这业务能力真没得说： 高难度动作、高密度协同、可落地的场景操作，放在整个春晚舞台上，都算得上独一份。 而这一套从感知、决策到协同的全链路智能操作，幕后操盘手，正是中国具身智能黑马公司—— 魔法原子 。 几年前，这类复杂机器人的展示还停留在实验室演示demo里，如今，他们把硬实力带到了国民舞台中央。 从技术验证到公众视野，中国具身智能产业，是真的开启「加速度」了。 从托马斯回旋到百台机器熊猫，硬实力拉满 而能稳稳完成这一系列操作，离不开Z1本身扎实的硬件底子。 在性能上，Z1采用自研关节组，拥有 24 个基础自由度，最高可扩展至 49 个，自由度覆盖 ...

Jay 发自 凹非寺 量子位 | 公众号 QbitAI 春节才是真正的大模型战场，全世界参与的那种。 大年初二，Anthropic史上最强Sonnet—— Claude Sonnet 4.6 发布。 话不多说，直接上视频。 不难看出， 计算机操作 是这次更新的主打卖点。 Anthropic表示，对填写复杂Excel、网页清单等任务，Sonnet 4.6 已经接近人类水平 。 其他方面也是全方位升级：编码、长上下文推理、Agent规划、知识型工作、设计……Beta阶段还支持 1M上下文 。 重点来了！ 定价依然跟Sonnet 4.5一样 ，免费用户也能用。 性价比简直高到离谱。 创业者Alex Finn体验后表示「难以置信」： 在大多数Agent任务上，Sonnet 4.6的表现跟Opus系列差不多好，速度还更快，价格只要1/5。 还不只一个人这么说。 Anthropic表示， 内测用户对Sonnet 4.6的喜爱程度，已经超过了超大杯Opus 4.5 。 史上最强Sonnet 计算机操作能力，可以说是这次Sonnet 4.6最亮眼的部分了，Anthropic也在这部分花了不少笔墨。 虽然跟最熟练的人类工作者比 ...

衡宇 发自 麦蒿寺 量子位 | 公众号 QbitAI 马斯克xAI人员大动荡，并没有妨碍它家新模型发布。 他在上对近十条Grok 4.2的夸夸推文又是点赞又是转发。 每一条都藏不住对自家新baby的认同和支持。 不仅如此，他还亲自发推公关： 公测将持续到下个月。公测结束后，Grok 4.2将比Grok 4快得多，也聪明得多。 我们知道目前仍有许多bug需要修复和改进，每天都在debug中～ 据了解， Grok 4.2的底层架构具备每周自我迭代的能力，以后每周将更新一次 。 风口浪尖上， Grok 4.2突然上线了——不过是公测Beta版 。 对比如今动辄数万亿参数的模型方阵，Grok 4.2的参数仅有500B，略显克制。 或许也是因为如此， Grok 4.2的市场和用户反馈呈现出一种诡异的两极分化 ：连连盛赞者亦有之，骂骂咧咧者有之。 面对那部分排山倒海的质疑声， 老马这位一向自信爆棚的硅谷狂人也有点坐不住 。 Grok 4.2公测版什么样？ 关于Grok 4.2，其实早有预告。 回顾Grok 4.2的诞生历程，可谓是一部标准的"鸽王进化史"。 去年12月起，马斯克就开始在上频繁预热 ，多次提到"3–4 周内 ...

Core Viewpoint - The article discusses the increasing presence and capabilities of humanoid robots in entertainment, particularly during the Spring Festival Gala, highlighting their potential for emotional value and educational applications in the consumer market [1][2][3][4]. Group 1: Humanoid Robots in Entertainment - The Spring Festival Gala featured various robots, showcasing their advanced capabilities and entertaining performances, which have evolved significantly from previous years [3][4][5]. - A standout performance was by a robot named "Little Bumi," which is a compact humanoid robot priced at 9,998 yuan, designed for consumer interaction and entertainment [24][25][38]. - The collaboration between Songyan Power and Volcano Engine resulted in a sophisticated voice interaction system for the robots, enhancing their ability to engage with audiences in real-time [42][46][62]. Group 2: Emotional and Educational Value - The founder of Songyan Power emphasized that the value of robots extends beyond practical tasks to include emotional support and companionship, particularly for children and the elderly [78][81]. - The article suggests that humanoid robots can play a significant role in K12 education, providing interactive learning experiences that traditional screens cannot offer [83][86][91]. - The demand for educational robots is strong among parents and schools, indicating a growing market for robots that can serve as teaching aids [91][96]. Group 3: Market Trends and Future Prospects - The article notes that the integration of AI in education is a significant trend, with humanoid robots expected to become essential teaching tools in the future [97][99]. - The potential market for educational robots is vast, with the possibility of widespread adoption beyond major cities [98]. - The development of a comprehensive ecosystem, including teacher training and curriculum design, is crucial for the successful implementation of robots in educational settings [99][100].