Sponsored by
Welcome back to The AI Field.
China's Moonshot just dropped Kimi K3, a 2.8 trillion parameter model that gives its weights away next week. CNBC reports the White House is now deciding who gets America's most powerful AI, which the White House denies. And starting today, Fable 5 is a standard part of Claude's Max and Team Premium plans. Three stories, one question: who's allowed to use frontier AI?
In today's AI Brief:
🌙 Kimi K3: China's open model reaches the frontier
🏛️ The White House now decides who gets frontier AI
🔓 Fable 5 becomes standard on Max plans today
Read time: 4 minutes.
Save 10+ Hours a Week With 37 Claude Prompts
Every manager faces the same situations before lunch: a message to land, a meeting to run, a hiring call, a report due. The AI Report built 37 Claude prompts for exactly those moments, organised by the situations every manager faces.
Copy the prompt, fill the brackets, run it in Claude, and get back 10+ hours a week. Oh, and it's free.
All you have to do is subscribe to The AI Report, a 5-minute daily AI brief read by 400,000+ business leaders at IBM, AWS and Microsoft, and the full prompt pack lands in your welcome email. The newsletter and the prompts, both free. Subscribe and grab both
LATEST DEVELOPMENTS
MOONSHOT AI
The AI Field: Moonshot released Kimi K3, a 2.8 trillion parameter model with native vision and a 1 million token context, and says the full weights go public by July 27. Their own benchmarks put it ahead of GPT-5.5, Opus 4.8, and GLM-5.2 on most coding and agent tests, while still trailing Fable 5 and GPT-5.6 Sol. That honesty is rare, and it makes the claims easier to believe.
Key details:
First open 3T-class model: 2.8T parameters with 16 of 896 experts active per token, built on Moonshot's new Kimi Delta Attention architecture.
In one 48-hour autonomous run, K3 designed and verified a working chip for serving a nano version of itself.
API pricing starts at $0.30 per million input tokens on cache hits, $15 per million out.
Available now in the Kimi app, Kimi Work, Kimi Code, and the API.
Why This Matters: Once the weights drop, near-frontier AI runs on anyone's hardware, no US lab subscription, no approval list. Every company doing the "open models do 90% for less" math just got a bigger number to plug in. The frontier is no longer a members-only club.
WHITE HOUSE
The AI Field: CNBC reports the Trump administration is dictating which companies and entities get access to new frontier models, a decision that used to belong to the labs. The White House denies it, insisting engagements are "voluntary." After June's forced Fable 5 shutdown, the leverage speaks for itself.
Key details:
Sources tell CNBC that future frontier rollouts will require explicit government approval of which partners are involved.
OpenAI gated GPT-5.6 to "trusted partners" at the government's request in June, before opening it up July 9.
Anthropic's restricted Mythos model already flows through the vetted Glasswing program, and OpenAI runs a similar one called Daybreak.
A White House official's denial: decisions on releases "rest entirely with the companies."
Why This Matters: Export controls used to stop at chips. Now they're climbing the stack into model access itself. If Washington holds the guest list for frontier AI, every roadmap in the industry has a new dependency it can't code around.
ANTHROPIC
The AI Field: Starting today, Fable 5 is included in every Max and Team Premium plan at 50% of usage limits. That ends three weeks of extensions, credits, and shifting access windows. Anthropic even apologized for the confusion.
Key details:
Included in all Max and Team Premium plans from July 20, at 50% of weekly limits.
Pro and Team Standard users keep access through usage credits, plus a one-time $100 credit.
Anthropic's explanation: demand "has been challenging to predict," with more capacity still coming online.
Why This Matters: You don't make your most powerful model a standard plan feature unless the compute finally arrived. The scarcity era of Fable 5 just ended. That's Anthropic saying it's ready to compete with GPT-5.6 on volume, not exclusivity.
THE AI FIELD'S TAKE
This week's theme is access. China's best lab is giving its frontier model away. Washington is reportedly deciding who gets America's. And Anthropic just turned its scarcest model into a plan feature.
Capability isn't the bottleneck anymore, permission is. Open weights answer to no one, US frontier models increasingly answer to the White House, and subscriptions answer to capacity. Three doors to the frontier, three very different doormen.
The question of 2026 isn't what the best model can do. It's who's allowed to use it.
QUICK HITS
📈 TSMC posted a record quarter and raised its outlook. Q2 revenue hit $40.2B, profit jumped 77%, and the CEO called AI demand "extremely robust" until around 2029-2030.
⚖️ In case you missed it: Apple sued OpenAI over trade secrets. The July 10 filing alleges job candidates brought Apple parts to interviews for "show and tell." The fallout is still spreading.
⌨️ OpenAI's first hardware drop is a $230 keyboard. Thirteen switches, built for driving Codex. Not the device Jony Ive was hired for.
💰 Nous Research is raising at a $1.5 billion valuation. Open agent models are getting unicorn-plus price tags now.
🏦 The Federal Reserve is talking about AI out loud. Governor Barr devoted a full speech to AI's economic effects. Central banks consider this their problem now.
AI TOOLBOX
🛠 ChatGPT Work - OpenAI's new agent that turns your notes and connected apps into finished decks, sheets, and reports. From the $20 Plus plan.
⚡ Grok 4.5 - SpaceXAI's "Opus-class" coding model at $2/$6 per million tokens, available in Cursor on all plans.
🧮 Leanstral 1.5 - Mistral's free open model that mathematically proves code correct, for software that can't fail.
WHAT TO WATCH
Kimi K3 weights, by July 27: the moment they're public, the "open models trail the frontier" argument gets its biggest stress test yet.
Gemini 3.5 Pro, third slip: the July 17 date came and went. Bloomberg says it missed internal quality goals, and prediction markets now lean late July to early August.
Fable 5 standard rollout: it starts today at 50% of limits. Watch whether that number survives contact with demand.
WRAP-UP
That's all for today's AI Brief.
🌙 Kimi K3: China's open model reaches the frontier
🏛️ The White House now decides who gets frontier AI
🔓 Fable 5 becomes standard on Max plans today
What did you think of today’s edition?
Until next time!
Olle | Founder of The AI Field