11月10日AI简报：Claude Fable 5.2 发布，Anthropic 拟募资 690 亿美元
一整天的大事全在这条里：Anthropic 一边发模型一边准备史上最大规模之一的IPO，Google 的 Gemini 4 Carbon 已经在 Vertex 上露头了。做开发的可以看看 Agents API 的12小时缓存和 Cursor/Copilot 里的新模型，关心价格的话 Muse Ember 每100万 token 才几毛钱。
Anthropic 于11月10日发布 Claude Fable 5.2 与 Mythos 5.2，速度较 5.1 提升约 30%，价格下降约 20%，其中 Mythos 仅对通过审核的组织开放。Fable 5.2 在 Artificial Analysis Intelligence Index 得分 65 居首，并在 CursorBench 以 60% 得分登顶。Anthropic 同时被曝计划以每股 70 至 80 美元在纳斯达克 IPO，股票代码 ANT，最高募资 690 亿美元。
各平台迅速跟进集成：Fable 5.2 已在 GitHub Copilot 全面可用，进入 Cursor 和 Perplexity Computer（Pro/Max 订阅）及 Search（Max 订阅）。Google 方面，Gemini 4 Carbon 已出现在 GCP SKU 目录和 Gemini 网页应用配置中，Vertex AI 上的 slug 已能响应，Logan Kilpatrick 发推预告发布；同时 Google 公布了用于训练的 TPU 8tx 和面向长上下文推理的 TPU 8i Max。Meta 上线了最小、最便宜的 Muse Ember 模型，定价为每百万 token 输入 0.10 美元、输出 0.50 美元，Muse 系列模型已登陆 Google Cloud 公测、Apple Watch 与 Wear OS。
OpenAI 的 Agents API 在 beta 中加入 12 小时 prompt 缓存、异步函数和按会话预算控制，ChatGPT Enterprise 套件上线 AWS Marketplace。NVIDIA 在 CoRL 发布了面向灵巧机器人的开放世界动作模型 GR00T 2 和 Isaac Lab 3.0。美团开源 1.6T 参数的多模态智能体模型 LongCat-2.5 权重，字节跳动的 Seedream 5.5 正在 Dreamina、CapCut 和 BytePlus 上线。
DAILY AI BRIEF 🗞 — Nov 10
MODEL FORECAST 👀: - Gemini 4 Carbon, today!!! - Gemini 4 Hydrogen (Flash), today? - GPT-6.2 Sol, this week? - GPT-6.2 Astra, this week? - Inkling 2 Small open weights, soon - Qwen4 Max/Flash/Plus/27B, this week? - Tencent Hy4 open weights and Flash, soon? - Grok 4.8, not soon? - Claude Fable 5.5 and Mythos 5.5, not soon? - DeepSeek V4.1 Pro, not soon?
ANTHROPIC 🔥: - Claude Fable 5.2 and Mythos 5.2 are out, about 30% faster and 20% cheaper than 5.1, with Mythos limited to vetted organizations. - Fable 5.2 tops the Artificial Analysis Intelligence Index at 65 (via ArtificialAnlys). - Ultra and Team Premium users get 1 to 3 Fable resets based on tenure, and Pro gets a week of Fable at up to half its weekly limit. - Anthropic is reportedly targeting $70 to $80 a share in a Nasdaq IPO under ANT, raising up to $69B.
OPENAI 🔥: - Dots can now connect to multiple local computers and pick which one runs each task. - ChatGPT Enterprise, with ChatGPT Work, Codex, and dots, is now on AWS Marketplace. - The Agents API added a 12-hour prompt cache, async functions, and per-session budget control in beta (via AzianMike).
GOOGLE 🔥: - Gemini 4 Carbon appeared in the GCP SKU catalog and Gemini web app config, and its slug already answers on Vertex AI (via AiBattle_). - Logan Kilpatrick teased the release by tweeting just "Gemini". - Google announced TPU 8tx for training and TPU 8i Max for long-context inference. - AI Studio can now build Google Maps Platform apps from a prompt, with 3D, Street View, weather, and routing. - Antigravity launched a Knowledge Work layout for research, writing, and artifacts without the terminal and git clutter. - Varun Mohan teased more Antigravity releases this week. - Gemini is rolling out Memory Summary to view and edit what it remembers about you (via SaadhJawwadh).
META 🔥: - Muse Ember, Meta's smallest and cheapest model at $0.10/$0.50 per 1M tokens, is live on the Meta Model API and Muse Code. - Muse Wonder, Muse Spark 1.3, and Muse Ember are in public preview on Google Cloud. - Muse is now on Apple Watch and Wear OS. - Meta is preparing to release Muse in the UK.
SPACEXSI 🔥: - Grok Bot weekly limits were reset for all users. - Grok Bot can now reach private MCP servers on your tailnet via Tailscale. - Grok Bot texting is getting closer, linking your number or Apple ID over iMessage only (via alexmichael). - Claude Fable 5.2 is in Cursor and tops CursorBench at 60% on max effort. - X launched Clipping on iOS and Android, with earnings shared between creator and clipper.
MICROSOFT 🔥: - Claude Fable 5.2 is generally available in GitHub Copilot.
PERPLEXITY 🔥: - Claude Fable 5.2 is in Perplexity Computer for Pro and Max, and in Search for Max.
NVIDIA 🔥: - NVIDIA unveiled GR00T 2, an open-world action model for dexterous robots, plus Isaac Lab 3.0 at CoRL.
BYTEDANCE 🔥: - Seedream 5.5 is rolling out on Dreamina, CapCut, and BytePlus.
MEITUAN 🔥: - LongCat-2.5 weights, a 1.6T-parameter multimodal agentic model, are out under MIT on Hugging Face.
ELEVENLABS 🔥: - ElevenAgents Campaign Manager runs outbound campaigns over phone and WhatsApp, following up with each contact until it hits a goal.
* Used Grok to compose this brief, cherry-picking the news and doing some post-editing. ** It feels a bit too much for Monday 💀
This report is categorized under AITOP's "Industry" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison and AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Claude, Gemini, Anthropic, OpenAI.
Source attribution: curated from TestingCatalog (@testingcatalog). Cross-source corroboration: 11 independent sources have covered this development (including Meta AI, Simon Willison’s Weblog, Artificial Analysis), confirming industry-wide signal strength.
