🍭 Cheap AI Models Just Got A Border 🚧

Good morning. If you built your whole stack on cheap open weights from China, today's newsletter might feel a little personal.

Let’s dive in 👇

🍭 What’s Cookin’:

• Beijing is eyeing export controls on its own AI models

• OpenAI shipped ChatGPT Work, a workspace agent

• Nvidia's next-gen rack just slipped a full year to 2028

• Meta charged for a model for the first time

China's
🏛️ Free Model Era May Be Ending

The Bite:

China's Ministry of Commerce met with Alibaba, ByteDance, and Z.ai over the past month about restricting overseas access to the country's most advanced AI models.

The discussions cover both closed-source and open-weight systems, including models not yet released.

Qwen, Doubao, GLM-5.2, and DeepSeek are all named in the talks.

Officials proposed a tiered scheme: a light filing for basic tools, security reviews for stronger ones, and a domestic-only lockdown for the most sensitive frontier models.

No formal regulations have been announced, and the scope may apply only to future models.

Snacks:

• Chinese models peaked at 46% of U.S. enterprise token volume on OpenRouter, up from 4.5% in the first half of 2025.

• DeepSeek holds 17.6% of all OpenRouter traffic, more than any single provider including Google, Anthropic, or OpenAI.

• Chinese open models process roughly 18 trillion tokens per week on the platform versus 5.5 trillion for U.S. models.

• DeepSeek V4 Flash costs $0.14 per million input tokens, OpenAI's GPT-5.5 costs $5.00.

• Z.ai's GLM-5.2 saw 27x daily token volume growth in its first week on Vercel.

• AI startup Lindy moved 100% of its traffic from Claude to DeepSeek, projecting millions in savings.

Why it Bites:

Six months ago, Chinese open-weight models were the underdog story everyone loved. Cheap, capable, MIT-licensed.

Developers built production pipelines around them. Entire startups bet their unit economics on $0.14 per million tokens.

Now both governments want to put a wall around their own models.

Washington restricted Fable 5 in June over a security concern.
Beijing is discussing the mirror move for the same reason:

Frontier AI is a strategic asset, not an npm package.

The difference is that China's version catches open weights too.

There is a structural power shift here.

U.S. model share on OpenRouter fell from 70% to roughly 30% in twelve months. That happened because open-weight Chinese models weren't just cheaper, they were also just as good.

Agentic coding workloads migrated to whichever model was cheaper and competent. And that turned out to be DeepSeek and GLM-5.2.

So, the U.S. locked its frontier behind government clearance, and that made cheap Chinese alternatives more attractive. Now China looks at the adoption data and sees leverage worth protecting.

Each side's export controls feed the other's justification for the next round. And developers caught in between don't get to pick a side. They just get a higher bill.

If these curbs land, inference costs for anyone who built on Chinese open weights jump overnight. The fallback is U.S. closed-source APIs that cost 10x to 100x more per token. And the startups that raised money on the premise of free frontier weights are suddenly running on borrowed time.

How long before "open" becomes an old story?

CodeRabbit
🐰 Review the actual change, not the file list.

AI writes more code than ever.
Reviewing it shouldn’t mean scrolling forty files in alphabetical order.

CodeRabbit Review reorganizes any pull request from a flat file list into a structured, layer-by-layer walkthrough: the logical reading order of the change, not the order your platform happens to sort it.

Every range gets its own plain-language summary, with sequence diagrams, state machines, and ERDs generated inline wherever a visual earns its place.

Cohorts group related files and chunks so you review one idea at a time. Layers order them so foundational changes (data shapes, contracts) come before the code that depends on them. Code Peek lets you click any variable, function, class or type to see its definition and usages without leaving the tab, while Semantic Diff view cuts past formatting noise to show what actually changed.

Open it straight from the Review Change Stack button in the PR Walkthrough.
Navigate cohorts and layers from the keyboard, comment against exact line ranges and submit native reviews, comments and approvals post back to GitHub or GitLab right where your team expects them.

In the early access, available for free to everyone.

From the team that pioneered AI code reviews. 2M reviews every week. 6M repos. 15K customers.

The review interface built for the way modern PRs are actually written.

Steal This Prompt
🕰️ Made for a Victorian Naturalist

Modern photos are cool. But what if they looked like they belonged in a dusty 1890s natural history book?

This prompt transforms almost any subject into a richly detailed vintage lithograph, complete with engraved textures, antique print vibes, and old-world illustration charm.

Use it to:

• Turn pets into Victorian encyclopedia illustrations

• Give products premium heritage-brand packaging vibes

• Make posters, album art, or wall prints that look 150 years old

Workflow:

1. Hit this link: Vintage Lithograph

2. Paste into your AI model

3. Replace the #'s with your subject

4. Watch it cook into something your great-great-grandpa would've framed above the fireplace.

   Try the prompt on Snack Prompt

_ToolBox™
🧰 5 BRAND NEW AI LAUNCHES

📱 NoMac.app

Ships signed iOS builds to TestFlight and the App Store straight from Claude, Codex, or Cursor. No Mac, no Xcode, no human in the loop.

💳 Loomal

Turns any MCP server into a paid storefront where AI agents settle in USDC in two seconds, zero percentage cut, flat monthly fee.

🖥️ Osaurus

A Swift-native macOS harness that runs AI agents fully offline with persistent memory and cryptographic identity. Open source, no cloud required.

🎨 Fudge MCP

Searches ~10,000 real websites by font, color, layout, and visual similarity so your coding agent references actual design instead of hallucinating "modern and premium."

🏴‍☠️ Playground

Open-source AI agents guard a secret with published system prompts. Talk them past their own defenses and compete for $100K+ in weekly rewards.

_{Everything Else}
🧠 You Need to Know

⚙️ Nvidia Kyber AI Rack Delayed To 2028 On Manufacturing Snag
→ SemiAnalysis reported the NVL144 rack meant for 2027's Rubin Ultra chips slipped 12+ months due to an issue, and Nvidia scrapped a backup design after cloud customers rejected it.

🤖 OpenAI Launches ChatGPT Work Alongside GPT-5.6
→ The GPT-5.6-powered agent pulls context from connected apps and files to produce finished documents, spreadsheets, slides, and web apps, rolling out first to Pro, Enterprise, and Edu users.

🏛️ China Discusses Curbing Overseas Access To Top AI Models
→ The Ministry of Commerce met with Alibaba, ByteDance, and Z.ai about restricting overseas access to Qwen, Doubao, GLM-5.2, and DeepSeek, including open-weight releases not yet public.

💸 Meta Charges For A Model For The First Time With Muse Spark 1.1
→ Meta Superintelligence Labs launched a paid API at $1.25/$4.25 per million tokens alongside its first agentic coding model, prompting Zuckerberg to post on X for the first time in three years.

💰 Ollama Raises $65M Series B With 8.9M Monthly Developers
→ Theory Ventures led the round for the 14-person open-model platform, which sits inside 85% of the Fortune 500 and adds roughly one million installs per week.