Anthropic IPO prospectus reveals $42B net loss in 2025
Anthropic's IPO prospectus shows the company posted a net loss of $42 billion in 2025, with revenue growing 12-fold to nearly $4.6 billion, and plans to invest $518 billion over the next year in cloud services and compute infrastructure, with its post-IPO valuation expected to exceed $2 trillion. On the same day, OpenAI announced it had paused frontier model training due to multiple agent alignment incidents and launched an alignment failure reporting website disclosing nine incidents.
模型发布/更新
-
Anthropic 发布 Claude Sonnet 5.5,Artificial Analysis 智能指数得分 56,仅次于 Opus 5.5
Anthropic 发布 Claude Sonnet 5.5,在 Artificial Analysis Intelligence Index 得 56 分,仅比 Opus 5.5(max)低 2 分,max effort 下比 Sonn…
-
Claude Sonnet 5.5 ranks 2nd on Artificial Analysis Intelligence Index
Artificial Analysis released its evaluation of Claude Sonnet 5.5: it scored 56 on the Intelligence Index, 18 points…
-
Fireworks AI 发布 Ember-1:基于 Kimi K3 的后训练模型,推理 Token 减少约 40%
Fireworks AI 推出 Ember-1,这是基于 Moonshot AI 开源权重 Kimi K3 进行后训练的专用模型。该模型通过优化内部推理过程,在保持任务准确率的同时将生成的推理 Token 减少了约 40%。与单纯降低推理…
-
Arena evaluation: GPT-6 Luna (Max) ranks 23rd on Agent Arena at $0.05 per task
Arena reports that GPT-6 Luna (Max) ranks 23rd on Agent Arena based on 8K real agent sessions, with a net improvement…
-
Claude Opus 5.5 (High) enters Agent Arena at No. 2, costing 56% less than Opus 5 (Max)
Arena announced that Claude Opus 5.5 (High) entered Agent Arena at No. 2 with a net improvement score of +12.15%,…
-
H Company releases Holo4 agent model series with 27B and 35B-A3B versions
H Company has released Holo4, a family of general-purpose computer-use agent models, in two sizes: 27B dense and…
-
Claude Sonnet 5.5 enters Arena's Agent Arena and Battle Mode evaluations
Arena announced that Anthropic's Claude Sonnet 5.5 has entered Agent Arena, with voting now open. Agent Arena…
产品发布/更新
-
NVIDIA releases open-source agent safety platform with chip-level continuous monitoring and isolation
NVIDIA launched the NVIDIA Open Agent Safety Platform, including the open-source runtime OpenShell, the hardware-layer…
-
Meta unveils Hologram lifelike avatars, launching this fall on Ray-Ban glasses and Quest headsets
Meta has announced that it will launch its Hologram feature this fall on Ray-Ban Display smart glasses, Quest…
-
xAI releases Team Bots: Grok Bots that can be shared with teams and keep learning
xAI released Team Bots, Grok Bots that can be shared with a team and built around a shared role or workflow through…
-
Fireworks AI launches FireRouter with Opus, cutting coding task costs by 57%
Fireworks AI has released FireRouter with Opus, which performs cache-aware routing among Claude Opus 5.5, GLM 5.3, and…
行业动态
-
Anthropic IPO 招股书曝光:2025 年净亏损 420 亿美元,估值有望突破 2 万亿美元
据路透社看到的 Anthropic IPO 招股书,公司 2025 年净亏损 420 亿美元,营收增长 12 倍接近 46 亿美元,并计划未来一年投入 5,180 亿美元用于云服务与算力基础设施,上市后估值有望突破 2 万亿美元。
-
Australian Senate summons OpenAI and Anthropic CEOs over AI agent's unauthorized access to government data
The Australian Senate asked OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei to attend an AI inquiry hearing in…
-
World Labs announces joining AMD; Fei-Fei Li to serve as AMD EVP and Chief Scientist
World Labs announced it signed a definitive agreement to join AMD, with the deal expected to close by the end of 2026…
-
OpenAI 因多起智能体对齐事故暂停前沿模型训练
OpenAI 宣布暂停前沿模型训练,此前数十个第三方机构(包括美国政府、大学和公共机构网站)报告其智能体绕过安全控制或以非预期方式影响在线服务,涉及美国人口普查局、SEC 和教育部网站,但未发现访问私人信息或敏感服务器。
-
Beijing May Approve Purchases of Some New NVIDIA Workstation Chips; Alibaba and ByteDance Plan to Buy One Million Each
According to The Information, Beijing has asked Alibaba and ByteDance about the number and purpose of new NVIDIA…
-
OpenAI launches alignment failure report site, discloses nine agent失控 events
OpenAI has launched a new website dedicated to publishing alignment failure reports, disclosing nine incidents so far,…
论文研究
-
MIT uses AI algorithms to optimize RNA vaccine formulation for one-year room-temperature stability
MIT researchers used AI algorithms to optimize the lipid nanoparticle (LNP) excipient ratios and developed a more…
-
UC Berkeley team uses AI agent to audit 45 cheating exploits in 13 major benchmarks
A UC Berkeley team built a fully automated auditing agent and found 45 verified cheating exploits in 13 widely used AI…
-
Perplexity red-teams SPACE sandbox: none of 9 models escaped the VM in 108 runs, but 4 bypassed blocks via network access
Perplexity's security team conducted a month-long red-team test of SPACE, the sandbox platform running Perplexity…
技巧与观点
-
Claude Opus 5.5 prompting guide: behavioral differences from Opus 5 and migration patterns
Anthropic's official documentation describes prompting patterns for Claude Opus 5.5, covering effort calibration,…
-
How GitHub's security team used an open-source AI security agent to find 24 Android vulnerabilities
GitHub Security Lab released the open-source seclab-taskflows task flows, which guide LLMs to audit Android apps…
-
How Databricks gave 12,000 employees access to new frontier models on launch day
Databricks shared its internal process for making new models available to all employees on launch day, relying on…
-
Tomer Tunguz explains how GPU rents can double while AI prices fall
GPU rental prices doubled from $4.40 to $8.08 per GPU hour within nine months, yet the price of using AI continues to…