Kimi K3: the largest open-weight AI model yet, and it's free
China's Moonshot AI has released Kimi K3, and the headline number is hard to ignore: 2.8 trillion parameters, making it the largest open-weight AI model ever shipped. More importantly for anyone choosing a model, it launched straight to the top of a major coding leaderboard — and the weights are free to download.
Kimi K3 is a mixture-of-experts (MoE) model — it has 896 "expert" sub-networks but only activates 16 per token, which is how it runs a 2.8-trillion-parameter model at a usable cost. It's natively multimodal (it understands images, not just text) and ships with a 1-million-token context window, enough to hold an entire codebase or a stack of documents in a single prompt.
How it stacks up
On the Arena WebDev leaderboard — a head-to-head ranking of front-end coding ability — Kimi K3 opened at number one, ahead of two of the strongest Western models:
| Model | WebDev score | Origin |
|---|---|---|
| Kimi K3 | 1,679 | Moonshot (open weight) |
| Claude Fable 5 | 1,631 | Anthropic |
| GPT-5.6 Sol | 1,618 | OpenAI |
On broader intelligence benchmarks it sits a notch lower — Artificial Analysis ranks it around 4th overall, behind Claude Fable 5 and the top GPT-5.6 reasoning settings — but for an open-weight model to trade blows with the best closed models at all is the story here.
Why "open weight" matters
Most frontier models (Claude, GPT, Gemini) are closed — you can only use them through the vendor's API. Kimi K3's weights are being released openly, so any team can download the model, run it on their own hardware, fine-tune it, and control their costs and privacy. That's a big deal for companies that can't send data to a third party, and it's a sign that the open-weight frontier — much of it now coming out of China — is closing the gap with the closed labs faster than expected. API pricing (for those who don't want to self-host) sits at $3 per million input tokens and $15 per million output.
The bottom line
Kimi K3 won't replace your everyday assistant overnight, but it resets expectations for what open models can do. If you care about self-hosting, data privacy, or simply avoiding lock-in, it's now a genuine option at the frontier — not a compromise. See where it fits on our Kimi page.