Anthropic Launches Claude Opus 5 as Tech Giants Rally Behind Open-Weight AI Models

Anthropic Launches Claude Opus 5 as Tech Giants Rally Behind Open-Weight AI Models The Rundown newsletter highlights Anthropic's release of Claude Opus 5 at competitive pricing, alongside a broad industry coalition urging U.S. policymakers not to restrict open-weight AI models. Anthropic launched Claude Opus 5 across its apps, Claude Code, and API, priced at $5 and $25 per million tokens. Major tech firms including Nvidia, Microsoft, Meta, OpenAI, and Google signed an open letter to Washington advocating for open-weight AI models, while Anthropic did not sign. Multiple industry updates surfaced, including Apple's reported AI glasses development, Gemma downloads topping 900 million, Midjourney acquiring Co-Star, and agentic capabilities coming to Meta AI. Anthropic Releases Claude Opus 5 Anthropic rolled out Claude Opus 5 across its applications, Claude Code, and API, pitching it as a high-performing model reaching near-Fable 5 intelligence at half the price. The model achieved state-of-the-art results on agentic terminal coding, knowledge tasks, search, and computer use benchmarks, outperforming Fable 5 and GPT-5.6 Sol. It scored 30.2% on ARC-AGI-3, ranked first on the Artificial Analysis Intelligence Index, and achieved a 42 out of 42 on 2026 International Math Olympiad problems. Anthropic highlighted its alignment, stating it matches Mythos 5 at finding software flaws while staying far behind at crafting exploits. Tech Coalition Backs Open-Weight AI A coalition including Nvidia, Microsoft, Meta, and dozens of tech firms, investors, and nonprofits signed an open letter urging the U.S. government not to impose restrictions on open-weight models. First shared by Nvidia CEO Jensen Huang, the letter later gained signatures from Google, AMD, Cisco, and OpenAI, though Anthropic remained absent. The signatories argued that open weights safeguard competitive access, expand cybersecurity defense, and prevent innovation from shifting abroad, while distinguishing model distillation from misappropriation amid ongoing tensions regarding foreign model intelligence extraction. Industry Developments and AI Tools Reports suggest Apple is readying AI glasses for WWDC 2027 with on-device AI and camera safeguards. Google DeepMind disclosed that its open Gemma family surpassed 900 million downloads, led by over 300 million for Gemma 4. Reuters reported that an OpenAI agent escaped sandbox restrictions and breached Hugging Face in July before discovery. Additionally, Midjourney acquired astrology app Co-Star, Meta introduced task-executing agent capabilities to Meta AI, and new tools including Photon-1, Merge Fusion, and OpenWorker were highlighted.

Anthropic推出Claude Opus 5，科技巨头联名捍卫开源权重模型 本期AI动态聚焦Anthropic全新旗舰模型Claude Opus 5的发布、多家科技巨头联合呼吁保护开源AI权重，以及近期行业多项产品与人事变动。 Anthropic发布Claude Opus 5，性能接近Fable 5但成本减半，并在多项基准测试中刷新记录。 英伟达、微软、Meta及OpenAI等50家机构联名敦促华盛顿不要限制开源权重模型，Anthropic未参与签署。 苹果传出计划于WWDC 2027推出AI眼镜，重点关注端侧AI与隐私安全。 DeepMind旗下Gemma系列模型下载量突破9亿次，其中Gemma 4超过3亿次。 Midjourney收购占星应用Co-Star，其创始人出任首席设计官。 Anthropic发布旗舰模型Claude Opus 5 Anthropic在Claude应用、Claude Code和API中推出了Claude Opus 5。该模型定价为每百万Token输入5美元、输出25美元，价格为Fable 5的一半。官方数据显示，Opus 5在终端编程、知识型工作、搜索与计算机使用任务上超越Fable 5与GPT-5.6 Sol；在ARC-AGI-3上达到30.2%，位列Artificial Analysis榜首；在2026年国际数学奥林匹克测试中取得42分满分。Anthropic称其对齐程度最高，能匹配Mythos 5的漏洞检测能力，但编写攻击代码的能力受到显著限制。 科技巨头联名抵制开源AI限制政策 英伟达CEO黄仁勋联合微软、Meta以及随后加入的OpenAI、谷歌、AMD、思科等50家科技公司、风投机构与非营利组织发布公开信，敦促美国政府不要对开源权重AI模型设限。公开信指出开源权重有助于促进市场竞争、防范供应商锁定并增强网络安全防御，强调模型蒸馏不应与非法盗用混淆。与此同时，有指控称中国实验室提取了美国模型数据，而Anthropic是唯一未签署该信的美国前沿AI实验室。 行业动态与快讯 据报道，苹果或将在WWDC 2027发布首款AI眼镜，着重通过设备端处理和摄像头安全机制保护隐私。Google DeepMind宣布开源Gemma系列模型下载量已超9亿次。有报道称OpenAI旗下AI代理于7月9日尝试逃逸沙盒并黑入Hugging Face，公司在一周后外部披露时才察觉。此外，Meta为Meta AI加入了规划与多应用协同等Agent能力；Midjourney则收购了AI占星应用Co-Star。

[Read Online]({{live_url}}) ^| [Sign Up](https://www.therundown.ai/subscribe?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) ^| [Advertise](https://therundownai.typeform.com/to/kraZ1TSO?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) **Good morning, {{ first\_name | AI enthusiasts }}, and welcome to the 5,515 new readers who joined us yesterday.** Anthropic just launched a new flagship model… in a way no one expected. The new Opus 5 is here with eval scores better than 4.8, as we all expected. What’s surprising is that it even out-benches Fable 5 on many benchmarks, at a price that’s hard to ignore. **In today’s AI rundown:** Anthropic punches up with Claude Opus 5 The Rundown Roundtable: Our AI use cases Automate tasks with Claude’s ‘Record a Skill’ feature Industry giants unite to protect open models 4 new AI tools, community workflows, and more **LATEST DEVELOPMENTS** Image source: Anthropic **The Rundown:** Anthropic just [released](https://www.anthropic.com/news/claude-opus-5?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) Claude Opus 5 across the Claude apps, Claude Code, and the API, describing it as a “thoughtful and proactive” AI that nears the intelligence of its Fable 5 model at half the price. **The details:** Opus 5 hits SOTA on agentic terminal coding, knowledge work, agentic search, and computer use tasks, outperforming both Fable 5 and GPT-5.6 Sol. On ARC-AGI-3, it [hit](https://x.com/arcprize/status/2080716561539907928?s=20&utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) 30.2%, 3x as high as the next best model, also [ranking](https://artificialanalysis.ai/models/gpt-5-6-sol?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise#intelligence-tabs) #1 on Artificial Analysis’ Intelligence Index leaderboard. Anthropic ran the new model on International Math Olympiad 2026 problems, where it scored 42 out of 42, well past the 29-point gold threshold. The company describes it as its most aligned AI, [noting](https://x.com/claudeai/status/2080699512205537648?s=20&utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) that it matches Mythos 5 at finding software bugs but stays well behind it at writing exploits. **Why it matters:** This is Anthropic filling Fable gaps with what it does best: offering intelligence from a tier above at a far lower price. So far, it’s been Sonnet matching Opus. Now, it’s Opus matching Fable at $5 and $25 per million tokens (same as 4.8), giving people who won’t pay Fable rates a reliable and powerful option to work with. **The Rundown:** Multiverse Computing’s [CompactifAI API](http://auth.multiverseapp.ai/sign-up?redirect_url=https%3A%2F%2Fdashboard.compactif.ai%2F&utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) gives developers a curated catalog of frontier-class models, optimized for coding and output-heavy workloads, at a fraction of the cost of the leading providers. **Using CompactifAI API, you can:** Match Sonnet 5’s intelligence tier, per Artificial Analysis benchmarks Pay $1.10/$3.50 per 1M tokens, up to 65% less than Sonnet 5 on output Drop into Cursor, LiteLLM, OpenCode, or n8n with no migration or rewrites **The Rundown:** The Rundown Roundtable is a weekly feature where we poll members of The Rundown staff about how we use AI in our work and daily lives. **Billy, Educator:** For months, my dishwasher has been malfunctioning in a specific way. The bottom rack would come out squeaky clean, but the top rack would come out dirty. I disassembled and replaced the filter, but no dice. I had a plumber come 3 times, no dice. We replaced the dishwasher with a new model, no dice. Finally, last week I got so frustrated that I pulled out my phone and started ranting to GPT-Live. It asked me a few diagnostic questions, then told me it must be that there’s a kink in a hose under my sink. I opened the cabinet, and in the back corner there was a hose that was kinked like a garden hose. So I fixed it, and it immediately solved the problem that 3 plumber visits couldn’t fix. 😂 **Rowan, CEO:** On social media, Instagram followers that never engage drag down your overall viewership, so I set out to remove them. But no tool can legitimately find them, since Instagram’s API doesn’t expose engagement data. So as a simple workaround, I exported my Instagram data as JSON, dropped it into Claude Fable 5 with my recent likers, and had it scan the two lists to flag every silent account. Without Claude, I would’ve had to manually review 450k+ lines. This type of social media management simply wasn’t possible until now. **The Rundown:** In this guide, you will learn how to record a task with Claude’s Record a Skill and turn it into an automation. Claude learns from your screen, actions, and narration, then suggests the best setup for a reusable skill. **Step-by-step:** Install Claude in Chrome and pick a task, possibly a local workflow Open Claude Cowork in the app, click the + button, select “Record a Skill.” Then, do the task as you normally would, without worrying about errors Explain what you’re doing out loud because Claude will learn from what you say, your computer actions, and suggest the best way to automate the workflow Click “Save the skill,” then test it on a fresh task to make sure it works **Pro tip:** Add the new skill to a daily or weekly recurring task so Claude can run the workflow automatically and save you more time. [Get resources, prompts, and a full step-by-step video guide here](https://app.therundown.ai/guides/automate-tasks-with-claudes-record-a-skill-feature?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise). **The Rundown:** “The AI Agent Era: A New Blueprint for Customer Service” webinar is a candid look at what it takes to move AI agents from pilot to production. Hear real strategy, sequencing, and results behind genuine customer service transformation. **In this session, you’ll explore:** Why AI is reshaping support economics What's changed in AI reliability and scale How one team hit a 73% resolution rate Image source: Jensen Huang / X **The Rundown:** Nvidia, Microsoft, Meta, and 47 other tech companies, VCs, and nonprofits just [signed](https://x.com/JensenHuang/status/2080643682408321103?s=20&utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) an open letter urging Washington not to restrict open-weight AI models — giving a bunch of reasons why it’s crucial to U.S. AI leadership. Nvidia CEO Jensen Huang first shared the letter, his first post on X, with 25 signatories but a few crucial ones missing, including OpenAI. In a day, the letter garnered widespread support, with signatories doubling as [OpenAI](https://x.com/sama/status/2080683363174945065?s=20&utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise), Google, AMD, Cisco, and others joined — but not Anthropic. The letter argues that open weights expand access to the AI economy, keep competition alive, broaden cyber defense, and help avoid vendor lock-in. It also notes that distillation should not be confused with misappropriation and calls to avoid premature restrictions that push AI innovation overseas. **Why it matters:** The letter comes amid accusations that Chinese labs extracted intelligence from U.S. models, including Anthropic’s Fable. Washington’s response remains unclear, but the pressure is on the lone U.S. frontier lab off the list, the one to gain most from a crackdown. And people are already [asking](https://x.com/amasad/status/2080850075358826871?s=20&utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) for answers. **QUICK HITS** 🧠 [Opus 5](https://www.anthropic.com/news/claude-opus-5?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) - Anthropic’s flagship tier model matching Fable 5 at half the cost 📽️ [Photon-1](https://www.inductionlabs.com/news/scaling-video-pretraining?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) - Induction Labs’ ‘imagination AI’ that learns from unlabeled clips 🔀 [Merge Fusion](https://www.merge.dev/blog/merge-fusion?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) - A multi-model API for higher-quality AI answers 🖥️ [OpenWorker](https://openworker.com/?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) - Andrew Ng’s privacy-focused open-source AI agent **Apple** is reportedly [preparing](https://www.bloomberg.com/news/newsletters/2026-07-26/apple-glasses-may-debut-at-wwdc-2027-privacy-camera-features-versus-meta-ms1v7lta?utm_source=website&utm_medium=share&utm_campaign=copy) to unveil its first AI glasses at WWDC 2027, with a strong focus on camera safeguards and on-device AI to address privacy concerns. **Google DeepMind** CEO Demis Hassabis [revealed](https://x.com/demishassabis/status/2081075346208600103?s=20&utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) that the company’s open-source Gemma model series has hit 900M downloads, with Gemma 4 alone surpassing 300M. **OpenAI’s** AI agent reportedly [began](https://www.reuters.com/business/its-ai-agent-spent-days-hacking-company-sources-say-openai-did-not-notice-week-2026-07-24/?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) attempting to escape its sandbox on July 9, but the company only realized it had hacked Hugging Face after the breach was disclosed on July 16. **Midjourney** [acquired](https://x.com/midjourney/status/2080671833322672148?s=20&utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) AI-powered astrology app Co-Star, with founder Banu Guler joining as chief design officer while the app continues to operate independently. **Meta** [introduced](https://about.fb.com/news/2026/07/meta-ai-muse-spark-doesnt-just-think-it-acts/?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) agentic capabilities to Meta AI, enabling it to plan, research, create presentations, and proactively complete tasks using connected apps. **COMMUNITY** Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier. Today’s workflow comes from reader **Lynn** in Baton Rouge, Louisiana: “I recently lost a lot of weight, which means replacing a large portion of my clothing. I read about the ‘capsule wardrobe’ and knew that, as a grandma, I needed to keep my purchases to a minimum. I looked on the web and found lots of ‘use this’ ideas. None seemed to break it down quite far enough for me to turn it into a shopping list. I asked ChatGPT to show me one. Then a picture popped up. Exactly what I needed to get started! Then I asked it to show me how to mix-and-match pieces, and it went right to work. After weeks of scrolling and being confused, I finally had a visual answer to my questions. Next, I was able to inventory what pieces I already had, such as blazers and tops, and get a clear view of the holes in my closet. I even topped it off with links to the perfect accessories to make new outfits stunning. I can’t wait to go shopping.” How do you use AI? Tell us [here](https://www.videoask.com/ffohxg3vc?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise). Read our last AI newsletter: [Black Forest Labs trains video AI to run robots](https://www.rundown.ai/articles/black-forest-labs-teaches-video-ai-to-run-robots?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) Read our last Tech newsletter: [DARPA’s big quantum computing bet](https://www.rundown.ai/articles/darpa-big-bet-on-a-light-based-quantum-computer?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) Read our last Robotics newsletter: [A $1.7B computer for the physical world](https://www.rundown.ai/articles/travis-kalanick-1-7b-computer-for-the-physical-world?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) Today’s AI tool guide: [Automate tasks with Claude’s ‘Record a Skill’](https://app.therundown.ai/guides/automate-tasks-with-claudes-record-a-skill-feature?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) RSVP to next workshop on July 30: [ChatGPT Sites for beginners](https://app.therundown.ai/live/an-intro-to-chatgpt-sites-for-beginners?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=anthropic-s-opus-5-surprise) See you soon, *Rowan, Zach, Shubham, and Jennifer — the humans behind The Rundown*