Mira Murati raised the largest seed round in history, $2 billion at a $12 billion valuation, spent 17 months training a 975 billion parameter model, and just released it as a free download under Apache 2.0.
The launch post even admits it trails the frontier. Let me explain why that might be the smartest move in AI right now.
Six months ago this company looked finished. The $50 billion follow-on round collapsed in January. Two cofounders went back to OpenAI. Zuckerberg tried to buy the whole thing, got told no, and poached a third cofounder instead.
The standard play from there is grinding toward a benchmark win to relaunch the fundraise. Murati went the other way. Nvidia investment in March plus a gigawatt of Vera Rubin compute. A Google deal in April. Then Inkling: 975B total parameters, 41B active, trained on 45 trillion tokens of text, image, audio, and video, with a 1M token context window.
The economics only make sense once you see what she's actually selling. Closed labs sell intelligence by the token. Thinking Machines sells Tinker, the platform where you fine-tune Inkling on your own data. The model is the free sample. The training infrastructure is the store.
That's why the honest line in the launch post is doing real work. A frontier model competes on raw capability. An open base competes on what a hospital, a bank, or a defense contractor can turn it into, and a customized 41B-active model that knows your codebase will outrun a genius general model that doesn't, at a fraction of the inference cost.
There's a supply story underneath too. For two years the strongest open weights came out of Hangzhou and Beijing: DeepSeek, Qwen, GLM. Every American enterprise that wanted to own its model was fine-tuning Chinese weights. Inkling hands them a US base scoring 77.6% on SWE-bench Verified.
Every lab launches by claiming state of the art. Murati launched by admitting she doesn't have it. The admission is the business model.
Enterprises will focus on in house open weights models trained on their own data for data privacy concerns
Frontier models will evolve into hardware may be
New Anthropic research: A global workspace in language models.
Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with.
We found a strikingly similar divide inside Claude.
One of the crisis of social media(may be) or generally in masses is how we use terms we have no deep idea about and use it so casually in conversations without knowing its true meaning
This in turn disregards that word completely even where it might be being used in true terms