The Biggest Open AI Model Is Now Free to Keep
Moonshot AI just released the world's largest open AI model — and unlike GPT or Claude, you can run it yourself, fine-tune it, and use it commercially.
The short version
A Chinese AI lab called Moonshot AI just made a very big model freely available to the world. It's called Kimi K3, and as of July 27th anyone can download it, run it on their own servers, and build products with it — no monthly contract with OpenAI, no usage fees to Anthropic.
To put the size in perspective: most of the models powering the tools you use today have a few billion parameters. Kimi K3 has 2.8 trillion. It's not just big for the sake of it — in independent benchmarks it ranked above Claude Opus-tier models on coding tasks.
Why this matters for your business
Until now, building something serious with frontier AI meant you were renting intelligence from one of two or three American companies. That's changing. With a model this capable available under an open licence, a small technical team can build you a custom tool — a client intake agent, an internal knowledge assistant, a product recommender — that runs entirely on infrastructure you control.
No per-query fees that explode when you scale. No data leaving your building if that matters to you. No vendor deciding to change the pricing next quarter.
It won't be the right choice for every project. Running a model this large still requires real hardware. But the ceiling just got a lot higher for what's possible without a big-tech contract.
Quick glossary
Open-weight model — An AI model whose inner workings have been made public, so anyone can download and run it. Like open-source software, but for AI.
Parameters — The knobs and dials inside an AI model that determine how it thinks. More parameters generally means more capability — and more computing power to run.
Fine-tuning — Taking an existing model and training it further on your own data, so it learns the specifics of your business, your tone, your products.
Context window — How much information the model can hold in mind at once. A 1-million-token window means it can read something roughly the length of ten novels before forgetting the beginning.
If you've been waiting to see whether open AI models could genuinely compete with the paid ones, this is probably the moment worth paying attention to.