Plus: Claude Opus 5 undercuts the frontier, Microsoft ditches the OpenAI tax, and an $8 chip runs an LLM
Sponsored by
Stop chasing leads, start converting them. Build smarter forms with Typeform AI and turn customer responses into automated workflows that move your business forward.
Try it freeToday, we've got a Chinese lab shipping a 2.8-trillion-parameter open-weight model that Wall Street couldn't stop talking about, Anthropic quietly halving the price of frontier-class coding, Microsoft admitting the future of AI is boring and cheap, and a language model running on a microcontroller that costs less than your lunch.
The theme this week is uncomfortable for a lot of people in San Francisco: the frontier is getting commoditized faster than anyone budgeted for. When the best model and the cheapest model are separated by weeks instead of years, the moat stops being capability and starts being distribution.
🐉 China Delivers a One-Two Punch to America's AI Lead
Moonshot AI unveiled Kimi K3 — a 2.8-trillion-parameter system it calls the world's largest open-source model — and Alibaba followed days later with a preview of Qwen3.8 at 2.4 trillion parameters. Moonshot's own testing puts K3 ahead of nearly every US system except OpenAI's GPT-5.6 Sol and Anthropic's Claude Fable 5, with full weights due to drop July 27. The strategic difference matters more than the benchmarks: while US labs lock their best work behind APIs, China is handing it out for free.
⚡ Anthropic Ships Claude Opus 5 at Half the Cost
Opus 5 landed Friday with what Anthropic says is nearly all of Claude Fable 5's intelligence for half the price — $5 per million input tokens and $25 per million output. It's now the default on Claude Max and the strongest model on Claude Pro. Buried in the release notes is the detail agent developers actually care about: mid-conversation tool changes no longer invalidate the prompt cache.
💸 Microsoft's In-House Models Cut Costs Up to 89%
Satya Nadella published a manifesto titled "Frontier Diffusion & Control" and announced Microsoft is routing traffic on its first-party surfaces to its own MAI models whenever they match frontier alternatives. Seven years and $13 billion after betting on OpenAI, Microsoft's position is now explicit: partners build the frontier, Microsoft makes it ordinary — and profitable.
🔌 Etched Doubles to $10.3B in Seven Months
The Harvard-dropout inference chip startup closed a $300M Series C led by Sequoia — the highest valuation ever for a Sequoia-led Series C — with a16z, SK Hynix, and Jane Street piling in. Etched says it's already booked $1 billion in orders and that demand keeps outpacing supply as customers move from evaluation to deployment.
🔬 esp32-ai squeezes a 28.9M parameter language model onto an $8 microcontroller — a delightful counterpoint to the week's trillion-parameter arms race, and a genuinely useful reference for anyone doing on-device inference.
🦀 Ruff just jumped from 59 to 413 default lint rules, making Astral's Rust-powered Python linter dramatically more opinionated out of the box — which matters a lot when an agent is writing most of your Python.
Coming in Hot
The AI answering service that answers 24/7 for small businesses.
Create cinematic 30-second AI videos with 4K output and multimodal references.
AI Infographic Generator
AI meeting assistant that records, transcribes, and acts
AI-powered influencer marketing analytics platform
All-in-one AI video, image, and music studio
AI-powered social media content creation and scheduling
AI content creation made simple
AI logo maker and brand identity platform
AI-powered cloud color grading and collaboration
Submit your AI tool to the #1 AI tools marketplace.
What it does: Ringing.io is an AI answering service for small businesses that answers phone calls 24/7, collects caller details, handles common requests, captures appointment inquiries, and sends structured summaries so teams can follow up without listening to vague voicemails.
Why It's Worth Your Time
Pricing
What it does: GraphicInfo is an AI infographic generator that turns prompts, data, and reference materials into polished, editable infographics. It helps creators transform reports, ideas, timelines, comparisons, and educational content into visual stories with AI-generated layouts and an editable canvas.
Why It's Worth Your Time
Pricing
What it does: Seedance 2.5 is an advanced AI video generator that creates cinematic videos from text prompts, images, videos, and audio references. It focuses on longer 30-second video generation, native 4K output, stronger character consistency, and multi-reference control for ads, storytelling, product videos, and creative projects.
Why It's Worth Your Time
Pricing
🤝 Cognition Buys Poke Because Personality Is a Moat Now
The acquisition brings Poke's conversational style to Devin, reflecting a growing conviction that how an AI talks to you is as commercially important as the model underneath it.
🧬 Reid Hoffman's New Lab Is Betting Against Coding
Prentis, co-founded with Mark Pincus, is in talks to raise $100M on the thesis that automating routine computer work will soon eclipse coding as AI's biggest use case.
🛡️ Hugging Face CEO Demands "Radical Transparency"
"The first autonomous agent cyberattack is an unprecedented event. It deserves an unprecedented response," he wrote. Expect this to shape the next round of agent safety standards.
