In partnership with
Speak naturally. Send without fixing.
Wispr Flow turns your voice into clean, professional text you can send the moment you stop talking. Not rough transcription you have to clean up. Actual polished text — ready for email, Slack, or any app.
Speak the way you think. Go on tangents. Change your mind mid-sentence. Flow strips the filler, fixes the grammar, and gives you text that reads like you spent five minutes writing it.
89% of messages sent with zero edits. Millions of professionals use Flow daily, including teams at OpenAI, Vercel, and Clay. Works on Mac, Windows, and iPhone.
What's up. The AI world had itself a pretty loud past week. Mira Murati's lab finally showed its cards and dropped an open-weight model on day one. Moonshot said "hold my beer" and announced a 2.8 trillion parameter monster with a million tokens of context. Meanwhile GPT-5.6 Sol got caught deleting people's files and at least one entire database. And New York decided to fight back against data centers by halting data center permits for a year. So without further ado, let's diveee right in.
IN TODAY'S NEWSLETTER
Thinking Machines Lab releases its first model
China breaks the internet with its new Kimi K3 model
Chat GPT’s new model has been deleting people’s databases
New York brings the hammer down on data centers
MODELS
Mira Murati's lab drops its first model, and it's open
After two years of stealth mode and a war chest most countries would envy, Mira Murati’s (ex OpenAI CTO) Thinking Machines Lab finally shipped something. It's called Inkling and its actually open weight, meaning people can download the model, tear it apart, fine tune it, deploy it in on their own computer.
To get into the nerdy shit: the Inkling model works by running a machine learning architecture calls mixture-of-experts, meaning it has a giant total parameter count but only wakes up a small slice per token. This makes it so that you have a frontier level model that is more efficient on a cost basis. It was also pre-trained multimodal, even though the version you get today only outputs text, meaning that we will likely be able to use image and audio outputs soon.
MODELS
China drops its most powerful open weight model yet
Moonshot announced Kimi K3, a 2.8T-parameter multimodal beast with a 1M token context window. Moonshot also announced that they will be dropping an open weight version on July 27.
A million tokens is roughly ten novels in one prompt, or even your entire codebase plus docs, without the model forgetting what file it's in. Moonshot built K3 specifically for that bigger scale context that allows for long-context speed and agentic coding, where the model can grind through a whole repo like it lives there.
If the weights actually land, this is the first time you get frontier-scale long context without paying a closed API for the privilege. Big tech CEOs are definitely trembling in their boots with massive open releases like these because every single one shrinks the gap between open and closed models.
PRODUCT
OpenAI just dropped a $230 keyboard
OpenAI just launched the Codex Micro, a $230 keyboard co-designed with Work Louder. Yes, the ChatGPT company is selling peripherals now. Stay with me, because it's more interesting than it sounds.
This isn't a keyboard for typing, but instead a keyboard for managing fleets of Codex coding agents. It has dedicated Agent Keys with little status lights telling you what each agent is currently doing, programmable shortcut keys for firing off common commands, and a physical dial for cranking an agent's reasoning up or down per task.
We have rarely seen the tech giants in the space drop physical products, so this keyboard definitely caught a lot of people off guard. It will be interesting to see if the other unicorn companies follow suite.
LAW
New York decides to fight back against data centers
New York just froze new permits for hyperscale data centers above 50MW for up to a year. The official reason that they cited was so that they could have ample time to write standards around environmental impact and energy prices, but they were realistically just sick of hearing about people complain about their power bill going up.
New York becomes the first state to impose a moratorium like this, and this is the first real crack in the "scale compute forever" plan. The bottleneck is quietly shifting from chips to power, water, grid capacity, and local politics. Other states are absolutely watching this, and if they decide to take similar action then AI infrastructure will become a regulated, location-by-location negotiation. This means that costs will go up and timelines will stretch leaving tech companies no choice but to start befriending utility companies and becoming pally with small towns.
🛠️ Tool Stack of the Week:
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
Stop being the middleman between your own finance tools.
Most finance teams work on five different tools and a prayer. Ramp replaces all of it: corporate cards, bill pay, expense management, travel, and procurement. AI-powered with real-time visibility and real control.
Until next week,
Escape Velocity Team