Meta推出新AI模型，Anthropic进展显著
2026-10-12 08:08 至 2026-10-13 08:08 （北京时间）
产品
1. How Anthropic’s Dynamic Workflows Coordinate Claude Agents
At AI Engineer New York, Anthropic drew on lessons from Unix and Linux to explain how agents can tackle bigger projects without overwhelming developers.
来源：Latent Space
2. Anthropic tests its next model under Claude Opus 5.5
Claude Opus 5.5 appears to be routing some users to a new, unannounced model. Early outputs on X look remarkable. Is it Opus 5.6 or Fable 5.5?
来源：TestingCatalog
行业
1. 阶跃 Step 5 Preview 接入腾讯 WorkBuddy，独家限免 2 周！
10 月 12 日消息，腾讯旗下全场景 AI 办公工作台 WorkBuddy 已接入阶跃最新旗舰模型 Step 5 Preview。即日起，该模型在 WorkBuddy 独家限时免费开放两周。凭借推理与 Agent 能力，Step 5 Preview 将助力 WorkBuddy 用户处理深度研究、信息检索、数据分析、文档与 PPT 生成等办公任务，并支持多步骤复杂任务的持续执行。 Step 5 Preview 是阶跃发布的新一代旗舰基座
来源：雷锋网
2. DeepSeek拟领投前腾讯AI科学家创业项目，Winverse AI估值或达4亿美元
DeepSeek正将业务触角延伸至人形机器人领域。 10月12日，据彭博报道，DeepSeek已同意领投Winverse AI的种子轮融资，金额达数千万美元。报道援引知情人士透露， 这笔交易将使这家成立仅数月的初创公司估值达到约4亿美元。 美团旗下龙珠投资（Long-Z Investments）等机构也已作出投资承诺。 这将成为DeepSeek迄今对外部初创企业最大规模的投资之一。 此前，DeepSeek已投资宇树科技。2026年8月，
来源：华尔街见闻
3. Anthropic's Andrej Karpathy says "the external world has seen the preview" of what AI can actually do
Andrej Karpathy, co-founder of OpenAI and now at Anthropic, sees a widening knowledge gap in AI. About 75 percent of the world's population has barely touched LLMs. Roughly 5,000 people, meanwhile, have access to interna
来源：The Decoder
4. Meta Unveils New Series of Advanced A.I. Models
The new models, Muse Wonder and Muse Wonder Full, are meant to compete with cutting-edge A.I. models from rivals like Anthropic and OpenAI.
来源：NYT Technology
5. Cyber-attack like OpenAI Medicare breach ‘seems likely’ in UK, says minister
Kanishka Narayan, concerned by Australian incident, says British cyber defences could be bypassed by artificial intelligence agent UK politics live – latest updates An AI-driven cyber-attack similar to the OpenAI breach
来源：Guardian AI
6. White House reverses course and restores UK AI Safety Institute's access to frontier models
The White House has reversed course and restored the UK's AI Security Institute's early access to AI models from Anthropic, Google DeepMind, and OpenAI. In September, the Trump administration had told the companies to st
来源：The Decoder
7. Meta has some Wonder Full new AI models
Meta is making up for lost time in the AI race. After launching Muse Spark, the first model from its revamped Superintelligence Labs division, and buzzy AI agent Muse, it's now launching two new models it claims can go t
来源：The Verge AI
论文
1. Meet the Anthropic Engineer Keeping Its AI Agent Trustworthy
If you’ve ever wondered how Claude and similar AI agents can answer complex questions in seconds and where they get the information, you’re not alone. Gurmeet Saran works to find that answer every day as an engineer at A
来源：IEEE Spectrum AI
推特
1. Over the last few months, we’ve made significant progress in reducing th…
Over the last few months, we’ve made significant progress in reducing the effectiveness of distillation attacks against Claude. In this post, we explain how:
来源：X · Anthropic
2. Great to see what the Fin team has built with NVIDIA Nemotron 🙌 They fin…
Great to see what the Fin team has built with NVIDIA Nemotron 🙌 They fine-tuned Nemotron 3 Super and Nano for their new Fin Apex customer service models. Congrats on the launch! Fergal Reid: We’re pleased to announce the
来源：X · NVIDIA AI
3. RT Chubby♨️ Now I can say it: Had early access to @Meta’s new "Watermelo…
RT Chubby♨️ Now I can say it: Had early access to @Meta’s new "Watermelon" for the past few days. Trick questions didn’t fool it, it followed instructions to the word, and as a coding agent it wrote, ran and checked its
来源：X · Meta AI
4. Le Chonk cooked on visual grounding. Pinpoints exactly where things are…
Le Chonk cooked on visual grounding. Pinpoints exactly where things are in an image. 42% on Dense 200, ahead of even frontier closed models..
来源：X · Mistral
5. Introducing QwenGyre, bringing elastic RL to extreme-long-horizon (XLong…
Introducing QwenGyre, bringing elastic RL to extreme-long-horizon (XLong) agent training. ✨ It launches with two core designs: 🥳 - Elastic Scheduler: reallocates GPUs between rollout & training as demand shifts—live exec
来源：X · Qwen
6. Shared some notes on Python 3.15, using it with uv, and features from Py…
Shared some notes on Python 3.15, using it with uv, and features from Python 3.11 I'm excited to use now that 3.10 is no longer the minimum supported version for my own projects
来源：X · Simon Willison
7. HVLP (High Very Low Profile) is an electrodeposited copper foil engineer…
HVLP (High Very Low Profile) is an electrodeposited copper foil engineered with an extraordinarily smooth bonding surface on the side that attaches to the resin or dielectric layer, departing from the naturally rough sur
来源：X · SemiAnalysis
8. This is the best piece I’ve read to date on the Chinese data economy for…
This is the best piece I’ve read to date on the Chinese data economy for LLMs, and how it compares to the US counterparts (summarized by Mercor, Surge, and Handshake having the biggest network, SUPPLY, but also many comp
来源：X · Nathan Lambert
本篇在北京时间早上 08:08 生成，汇总此前 24 小时的报道和推特，标题直链原文。
