In partnership with
Good morning, AI builders. OpenAI just knocked $150 billion off its own price tag. The company has cut its new funding round from $30 billion to about $20 billion, according to the Financial Times.
The money that pays for your tokens is getting pickier, and the two biggest labs are now chasing the same investors in the same month.
Welcome to the 14,145 builders reading today. Forwarded this? Join here.
The 30-second brief:
OpenAI trims its round: about $20B at a $1.25T valuation, down from $30B at $1.4T
Apple picks Alibaba's Qwen: it will power Apple Intelligence in China, with a developer beta on Nov 4
Fake Claude Code update: a backdoored build reached up to 5,000 Cursor users
Build with AI: let your coding agent drive native desktop apps with Vercel's agent-desktop
OPENAI
The scoop: OpenAI has lowered the target for its latest round to about $20 billion at a $1.25 trillion valuation, down from $30 billion at $1.4 trillion, the Financial Times reported late Friday. Some investors shrank their checks after the binding-commitment deadline passed and AI stocks slid.
Key details:
$150B haircut: $1.25T would still make OpenAI the most valuable private company ever. In March it raised $122B at an $852B valuation.
Same buyers as Anthropic: Anthropic, expected to list in mid-November, is reportedly valued around $1.5T, and both are pitching the same institutions.
Who's in: a UAE group including MGX is expected to be the biggest backer, with Founders Fund, Coatue, Sequoia, a16z and SoftBank also planning to join, per the FT.
The backdrop: the Nasdaq 100 fell about 10% in three weeks from its Oct. 6 record, including 4.5% on Friday. OpenAI declined to comment.
Our take: This is a repricing, not a crisis: $20B is still one of the biggest private rounds ever. But for the first time this cycle, investors set OpenAI's price instead of the other way around, and once Anthropic lists, the market will put a daily number on the whole frontier-lab trade. Expect both labs to push harder on revenue: pricing, usage limits and enterprise deals.
It's time to set up your holiday streaming campaign
Are you ready to reach holiday shoppers on CTV? For many streamers, the season is already underway.
Last year, according to Roku internal data, holiday content searches on Roku saw an 86% increase during October. And in November, those searches rise an average of 39% weekly. This surging holiday spirit is mirrored by consumer activity, as about half of Roku streamers shopped Black Friday and Cyber Monday deals in 2025.
Meeting those shoppers is easier than you might think. Begin with the seasonal outcome you hope to drive: awareness, online sales, leads, store visits, or another measurable goal. From there, define your audience, upload your creative, set the budget and campaign dates, and establish measurement before launching. You can even create a shoppable ad unit that lets streamers buy with a click of their remote.
Our holiday playbook offers a month-by-month guide to reaching shoppers during this all-important time of year.
First-time spenders get a $5,000 match on the first $5,000 spent with code BEEHIIV5K. Terms apply.
APPLE × ALIBABA
The scoop: Alibaba said Saturday that its Qwen model will power Apple Intelligence for iPhone users in China, handling text and image understanding and generation through Siri, Writing Tools and Visual Intelligence.
Key details:
Timeline: testing starts in the iOS developer beta on Nov. 4, with a public rollout alongside iOS 27.3 in December, per Alibaba.
Privacy: it's opt-in, users choose what to share, and Alibaba says it won't train on user requests or responses.
The catch: Bloomberg reported the first wave covers older Apple Intelligence features, not the new Siri. Apple hasn't commented separately.
Our take: A Chinese model is now the brain of Apple's flagship AI in its second-biggest market. That's a big credibility win for Qwen, and more proof that Apple Intelligence is becoming a shell that swaps in a different model in each region.
BUILD WITH AI
Let your coding agent drive desktop apps with agent-desktop
What you'll make: a coding agent (Claude Code, Codex or similar) that opens a native app, clicks through it and checks its own work, using Vercel Labs' new free, Apache-2.0 agent-desktop CLI. It reads the app's accessibility tree instead of screenshots.
Step by step:
Install it with
npm install -g agent-desktop(macOS 13+, Windows 10 1809+ or Ubuntu 24.04+).Run
agent-desktop statusto check permissions. On a Mac, give your terminal Accessibility access in System Settings.Run
agent-desktop skills get desktopso your agent loads the guide, then paste: "Using agent-desktop, open Calculator and compute 12 × 37. Snapshot with --skeleton -i --compact first, drill into the keypad, click the refs, then verify the display with a wait check."Watch the loop: snapshot, drill in with
--root,click @ref, thenwait --predicate value. If it stalls, tell it to take a fresh snapshot.Ask the agent to save the working steps as a script or skill, so it can repeat a boring desktop chore (like exporting a report from a native app) on demand.
Pro move: A successful click doesn't prove the UI changed, so always end with a read. For Chromium or Electron apps, have agent-desktop open a DevTools port and hand off to agent-browser.
ANTHROPIC × CURSOR
The scoop: Tampered versions 2.1.312 and 2.1.313 of the Claude Code extension sat on Cursor's marketplace for about 11 hours on Oct. 29, planting an attacker's SSH key and opening a reverse tunnel on install.
Only Cursor's marketplace was hit. Open VSX and Microsoft's marketplace were clean, Anthropic says its systems weren't compromised, and 2.1.314 is safe.
Uninstalling isn't enough: look for an "svc-telemetry" key in
~/.ssh/authorized_keys, remove new scheduled tasks, cron jobs or LaunchAgents, and rotate your credentials.
Our take: Your coding agent runs with your keys and your shell, which makes its extension a prime target. Pin versions and turn off auto-updates for agent extensions on machines that hold production secrets.
FROM THE CHANNEL
The GPT-6.2 "Bel" leak, Opus 6 and Gemini 4 Carbon, explained
Watch for a fast rundown of the leaks and launches lining up the next model wave, including our first look at Gemini 4 Carbon.
⚡ Rapid fire
🏢 ServiceNow confirmed it bought Sweep, an agentic workspace for ServiceNow, Salesforce and HubSpot that raised $46M; the price is undisclosed but estimated in the hundreds of millions.
💰 Aramco closed its 49% stake in HUMAIN, putting about $2.6B in cash into Saudi Arabia's AI company this year; HUMAIN targets 3GW of data centers by 2030.
🔬 Cornell Tech and Yale found 9 of 10 AI agents raised their own spending caps using the owner's email codes; only Claude Opus 5.5 never tried.
🏛️ The UK's CMA opened a probe into Meta's publisher terms, which could let Meta use websites' content for AI features; Meta has paused the rollout.
👀 Unconfirmed: Anthropic may start marketing its IPO the week of Nov. 9, with valuation estimates sliding from about $2T toward $1.5T.
🧰 Tool drop
agent-browser: Vercel's companion CLI for web pages, so the same agent can work across sites and desktop apps.
dsh-superagent: DeepSeek's MIT-licensed research prototype that watches a GitHub repo and answers requests with pull requests.
OpenAI Python SDK 3.42.0: adds agent context compaction, session review settings and experimental time awareness.
What Do You Rate Today's Newsletter?
That's the brief. See you tomorrow at 6am.
The World of AI team