Chinese startup Moonshot AI drops massive 2.8T parameter Kimi K3 model for free
While Western tech giants lock their best AI behind expensive subscriptions, Moonshot AI is casually handing over the keys to a monster model. Open-source artificial intelligence just got ridiculously serious.
Chinese AI startup Moonshot AI scheduled the public weight release of its flagship Kimi K3 model on the Hugging Face platform. The model packs 2.8 trillion parameters alongside a 1-million-token context window, positioning it as the first open-source entry in the 3-trillion-class weight bracket.
Underneath the hood, the architecture relies on a sparse Mixture of Experts system that activates 16 out of 896 specialized experts per task. Combined with custom attention mechanisms, the setup delivers a 2.5-times boost in scaling efficiency over its predecessor, keeping hardware requirements from melting local data centers.
The company originally throttled user signups on its web platform to prevent server meltdowns during high demand. Moving the weights to the open-source community transfers the heavy compute burden directly onto external inference partners and independent maintainers who are preparing ecosystem support.
Giving away trillion-parameter blueprints forces the entire industry to question whether paid proprietary API access still makes economic sense. Open-source releases of this magnitude usually turn corporate boardrooms into panic rooms overnight.
Source: Hugging Face
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.