⏱ The 10-second version
Chinese open-weight models now handle 30 to 46% of US companies' AI traffic on OpenRouter, up from 11% a year ago. The new discipline: route every task to the cheapest model that clears the bar.
The White House is in advanced talks with OpenAI, Anthropic, and Google on voluntary frontier-model standards. An announcement could land as soon as next week.
Today's Two-Minute Win: The “Get paid more” prompt that turns vague salary negotiations into real business discussions.
The Big Thing: good enough just got a price tag
CNBC reported yesterday that US companies are quietly rerouting a huge share of their AI workloads to Chinese open-weight models. On OpenRouter, Chinese models have held above 30% of weekly token share since February, peaking at 46%. A year ago the average was 11%. Z.ai's GLM 5.2 landed within a point of Claude Opus 4.8 on a closely watched agentic benchmark at roughly a fifth of the cost, and Vercel says its daily token volume grew 27x in its first full week.
Why you care: frontier token prices are rising (Anthropic's Fable 5 moves to usage credits July 12th at $10/$50 per million tokens), and "tokenmaxxing" is officially out. Efficiency is the new flex.
What to say in your next meeting: "Which of our AI tasks actually need a frontier model? Everything else should route to the cheapest model that passes our quality bar. Let's find out what that bar is."
Source: CNBC
Speed Round
Norm AI raised $120M at a $1.2B valuation to automate legal and compliance work. Lawyers, your billable hours have entered the chat. Tech Startups
Together AI closed an $800M Series C led by Aramco Ventures at an $8.3B valuation, with Q2 bookings already past $1.15B. Open-source infra is a real business now. Crunchbase News
Microsoft launched "Frontier Company": $2.5B and 6,000 engineers to personally install AI inside enterprises. The models are commodities; the deployment is the product. TechCrunch
The Future of Life Institute's new AI Safety Index is out and nobody gets an A. The whole class is failing together, which is somehow not comforting. TIME
Microsoft is also cutting 4,800 jobs in the latest AI-cited restructuring, part of a running 2026 layoff tally that keeps name-checking AI. NBC News
The UN's new AI for Good Global Commission holds its first meeting today in Geneva, co-chaired by Marc Benioff and Rwandan President Paul Kagame. Tech Startups
The Two-Minute Win
Prompt of the Day
The “Get paid more” prompt
Most people ask for a raise with vibes.
Bad plan.
Your manager does not need a speech. They need a clean business case they can repeat when you are not in the room.
Paste this into Claude:
“Act like a career strategist and compensation coach.
Help me build a case for a raise or promotion based on the notes below.
Turn my messy work history into:
A one-sentence positioning statement
My 3 strongest value themes
A list of measurable wins, rewritten in business language
The strongest argument for why my scope has outgrown my current title or pay
A 60-second script I can say to my manager
A short follow-up email after the conversation
The objections my manager might raise, and how I should respond calmly
Rules:
Make it confident, not entitled.
Tie everything to business value.
Replace vague phrases with measurable impact.
If numbers are missing, show me where to add them.
Do not exaggerate.”
Then paste your messy notes: projects, dashboards, clients helped, revenue protected, time saved, costs reduced, problems solved, leadership moments, and anything you do now that was not in your original job description.
This turns a vague “I think I deserve more” into a raise case your manager can actually defend.
Estimated time saved: 45 minutes.
Potential upside: awkward conversation becomes an adult business discussion.
Sponsored (by us, shamelessly)
Two Minutes AI helps draft your deep-work tasks so you can review and ship instead of staring at a blank page. Early-adopter access to this productivity system is now open at twominutesai.app.
That’s it for today.
Thanks,
Don