🤖 AI 日报 | 今日十大热点（2026.04.26）

来源：Curious Spark · 每日 AI 精选

1. 🏆 谷歌宣布400亿美元重磅投资Anthropic，AI双寡头格局正式确立 — cnbc.com

谷歌正式宣布对Claude系列大模型开发商Anthropic追加投资，总规模高达400亿美元，其中超过60%以谷歌TPU算力资源形式划拨，为Anthropic未来三年前沿大模型训练锁定产能。此轮投资将Anthropic估值推升至3500亿美元，成为AI史上最大单笔定向投资。与此同时，亚马逊也将对Anthropic投资追加至250亿美元。分析人士指出，本次交易正式确立「微软+OpenAI」对阵「谷歌+Anthropic」的AI双寡头格局，未来Android系统将预装Claude入口，云端将打通Gemini与Claude双引擎接口，行业竞争格局加速重塑。

Google officially announced a $40B investment in Anthropic, with 60%+ delivered as TPU compute, locking in frontier model training capacity for 3 years. Anthropic’s valuation surged to $350B — the largest targeted AI investment in history. Amazon also raised its stake to $25B. Analysts say this formally establishes a Google+Anthropic vs. Microsoft+OpenAI duopoly, with Android to pre-install Claude and Google Cloud to bundle both Gemini and Claude APIs.

2. 🤖 马斯克xAI发布Grok 4.20：推理能力反超GPT-5.4，语音模型同步上线 — xai.com

马斯克旗下xAI正式发布Grok 4.20及Grok 4.3两个新版本，重点强化推理与科学计算能力，在多项学术基准测试中反超GPT-5.4，尤其在数学竞赛和代码生成维度表现突出。与此同时，xAI同步推出语音思考模型grok-voice-think-fast-1.0，支持实时多轮语音对话，直接进军智能语音助手赛道。随着SpaceX对xAI的战略并购整合，xAI已成为一家估值1.25万亿美元、融合太空与AI能力的超级实体，被市场视为2026年最具颠覆潜力的AI新势力。

xAI launched Grok 4.20 and Grok 4.3, with significantly improved reasoning and scientific computation that outperforms GPT-5.4 on multiple academic benchmarks, especially in math and coding. The new grok-voice-think-fast-1.0 voice model enables real-time multi-turn voice conversations. Following SpaceX’s acquisition, xAI is now a $1.25T entity combining space and AI, seen as one of 2026’s most disruptive forces.

3. 💡 OpenAI推出生命科学专属模型GPT-Rosalind，药物研发AI时代加速到来 — openai.com

OpenAI正式推出垂直领域专属模型GPT-Rosalind，面向生命科学和药物研发场景深度定制，能够自主完成蛋白质结构预测、候选药物筛选、临床试验数据分析等复杂科研任务，辅助效率较传统方法提升超过5倍。该模型整合了全球数百万篇生命科学论文的结构化知识，支持科研人员以自然语言完成跨学科知识检索与实验设计。与此同时，OpenAI内部确认新一代GPT-6预训练已完成，重点优化长期自主任务执行能力，预计年中正式发布，被业界寄望为迈向AGI的关键里程碑。

OpenAI launched GPT-Rosalind, a life-sciences-specific model that autonomously handles protein structure prediction, drug candidate screening, and clinical trial data analysis — over 5x more efficient than traditional methods. It integrates millions of structured biomedical papers for cross-disciplinary research. OpenAI also confirmed GPT-6 pretraining is complete, focused on long-horizon autonomous task execution, with a mid-year launch seen as a key AGI milestone.

4. 🚀 特斯拉Optimus Gen 3规模上岗：千台机器人工厂作业，售价目标2万美元 — tesla.com

特斯拉正式宣布Optimus Gen 3人形机器人已在Giga Texas工厂实现规模化上岗，超过1000台机器人与人类并肩完成零件分拣、质检和组件搬运等工序，标志着人形机器人从实验室走向真实产线的历史性节点。特斯拉透露目标年内将工厂机器人部署量扩大至10000台，并计划于工厂验证完成后向消费市场开放，目标售价2万至3万美元，有望成为史上首款大规模量产的通用人形机器人。马斯克表示，Optimus结合Dojo AI与FSD技术，未来将进入家庭场景，取代人工完成更广泛的体力劳动。

Tesla announced over 1,000 Optimus Gen 3 humanoid robots are now working on production lines at Giga Texas alongside humans, handling parts sorting, quality inspection, and component transport. Tesla targets 10,000 factory deployments this year, with consumer sales planned post-validation at a $20K–$30K price point — set to become the world’s first mass-produced general-purpose humanoid robot. Musk says home deployment is next.

5. 🎯 Boston Dynamics电动Atlas商用落地，IPO估值或达千亿美元 — bostondynamics.com

Boston Dynamics宣布全电动版Atlas人形机器人正式进入商业部署阶段，已在现代汽车工厂承担实际生产任务。新版Atlas拥有56个自由度、50公斤举升能力、IP67防尘防水等级，并支持自动换电，是目前商用场景中灵活性最强的人形机器人产品。与此同时，外媒披露Boston Dynamics正积极筹备IPO，市场估值预期高达1000亿美元，将成为机器人赛道规模最大的上市企业之一。分析师指出，人形机器人行业正式进入「比拼真实产线表现」阶段，现代汽车作为大股东的战略背书，将显著加速Atlas的全球商业化推广。

Boston Dynamics deployed its fully electric Atlas in commercial production at Hyundai plants, with 56 degrees of freedom, 50 kg lift capacity, IP67 rating, and autonomous battery swapping. The company is reportedly preparing an IPO with a projected $100B valuation, which would make it one of the largest robotics listings ever. Analysts say the humanoid robot industry has entered the ‘real production performance’ phase, with Hyundai’s backing accelerating global commercialization.

6. 📡 DeepSeek估值突破200亿美元，腾讯阿里争相入局战略融资 — 36kr.com

据多家媒体报道，深度求索（DeepSeek）首次启动大规模外部融资，目标估值已突破200亿美元，腾讯与阿里巴巴等巨头均已进入深度洽谈阶段。DeepSeek凭借V4系列开源模型横扫各项基准测试，并以极低的API定价颠覆国际巨头的商业逻辑，迅速成为全球AI产业最受瞩目的新势力。此次融资若成功落地，将是中国AI创业公司有史以来规模最大的单轮融资，也标志着国产开源大模型正式从「技术突破」迈向「资本扩张」的新阶段，对全球AI竞争格局将产生深远影响。

DeepSeek is reportedly seeking its first large-scale external funding round with a target valuation exceeding $20B, with Tencent and Alibaba in advanced talks. Riding the success of its V4 open-source models and disruptive pricing, DeepSeek has become the most-watched AI startup globally. If completed, this would be the largest single-round funding for a Chinese AI startup, signaling that domestic open-source LLMs are moving from technical breakthroughs to capital expansion.

7. 🧠 字节豆包月活破1.16亿，国产AI应用首次进入全球Top 3 — bytedance.com

字节跳动旗下AI助手豆包（Doubao）最新数据显示，C端月活跃用户已突破1.16亿，凭借抖音、今日头条等平台的流量优势，成为首款月活跃用户进入全球前三的国产AI应用。豆包在多模态创作、实时对话、娱乐内容生成等场景积累了大量年轻用户，低延迟与强大的文图生成能力是其核心竞争优势。分析师指出，豆包的崛起证明中国AI应用市场已形成独立的增长逻辑，字节凭借超大流量池将持续强化豆包与创作生态的深度绑定，向ChatGPT的全球统治地位发起有力冲击。

ByteDance’s AI assistant Doubao reached 116 million monthly active users, becoming the first Chinese AI app to break into the global top 3. Powered by ByteDance’s Douyin and Toutiao traffic, Doubao leads in multimodal creation, real-time conversation, and entertainment content generation among younger users. Analysts say this proves China’s AI app market has developed its own growth dynamic, with ByteDance’s traffic flywheel challenging ChatGPT’s global dominance.

8. 💰 Figure AI获390亿美元估值，人形机器人「劳动力即服务」商业模式成型 — figureai.com

人形机器人独角兽Figure AI完成C轮融资，估值达到390亿美元，OpenAI、微软、英伟达及杰夫·贝佐斯均参与战略投资。Figure最新一代机器人已在宝马斯巴达堡工厂正式上线，能够理解自然语言指令独立完成复杂装配任务，效率媲美熟练工人。Figure CEO表示，公司核心商业模式为「机器人劳动力租赁服务」（Workforce as a Service），按使用时长向制造业客户收费，有效降低企业一次性采购门槛。随着人形机器人技术加速成熟，2026年正在成为「机器人代替真实岗位」的元年，劳动力市场正面临前所未有的深层冲击。

Figure AI closed a C-round at a $39B valuation with OpenAI, Microsoft, NVIDIA, and Jeff Bezos as strategic investors. Its latest robot is live at BMW’s Spartanburg plant, completing complex assembly from natural language instructions at skilled-worker efficiency. Figure’s ‘Workforce as a Service’ model — charging manufacturers by usage hour — lowers adoption barriers. As humanoid robots hit production scale in 2026, labor markets face unprecedented disruption.

9. 🔬 阿里Qwen3开源模型全球下载突破6亿次，国产大模型生态影响力全面爆发 — qwenlm.github.io

阿里巴巴通义团队宣布，Qwen3系列开源大模型在全球各大平台累计下载量已突破6亿次，在Hugging Face开源模型排行榜上连续多周位居全球第一，标志着国产开源大模型的国际生态影响力达到历史新高。Qwen3旗舰模型在MMLU、HumanEval、MATH等主流基准测试中全面超越国际主流开源模型，并在代码生成、中文理解和多语言推理上建立显著优势。阿里同步宣布将进一步扩大开源投入，下一代模型将支持原生多模态输入和万亿级参数规模，持续巩固国产AI在全球开源生态中的核心地位。

Alibaba’s Qwen3 open-source model series surpassed 600 million downloads globally and topped the Hugging Face leaderboard for multiple consecutive weeks. The flagship Qwen3 model outperforms international open-source peers on MMLU, HumanEval, and MATH benchmarks, with strong leads in code generation, Chinese NLP, and multilingual reasoning. Alibaba plans to expand open-source investment, with the next generation targeting native multimodal input and trillion-parameter scale.

10. 🌐 Q1全球AI风投创纪录：2970亿美元融资中81%流入AI赛道，独角兽批量诞生 — kersai.com

最新风险投资数据显示，2026年第一季度全球创业公司融资总额高达2970亿美元，其中AI领域独揽2420亿美元，占比达到惊人的81%。OpenAI、Anthropic、xAI三家公司单轮融资规模均超百亿美元，吸走全球AI风投总量的绝大部分。AI原生浏览器、AI手机、AI代理平台、AI医疗等细分赛道涌现数十家新晋独角兽，估值普遍超过10亿美元。分析师警告，资本的极度集中正在加速行业洗牌，大量传统软件公司将在12至18个月内面临AI原生替代产品的生存冲击，就业市场结构性变革已不可逆转。

Q1 2026 saw a record $297B in global startup funding, with AI capturing $242B (81%). OpenAI, Anthropic, and xAI each raised $10B+ rounds, dominating the flow. AI-native browsers, phones, agent platforms, and healthcare AI spawned dozens of new unicorns valued above $1B. Analysts warn that this capital concentration is accelerating industry consolidation, with traditional software firms facing AI-native disruption within 12–18 months and structural job market changes becoming irreversible.

📌 关注 Curious Spark，每日追踪 AI 前沿动态
如有转载请注明来源