Kimi K3 Released — 2.8 Trillion Parameters, 1 Million Token Context, Model Will Be Open-Sourced on July 27 🔆

We tested it yesterday and share details: the model is on par with Opus 4.8 and GPT-5.5.

We are increasingly considering switching to Kimi. The main thing is that it's an open model — you can download and run it on your own server. Start working through the cloud, and if your business grows or you have confidential data, you can easily migrate to your own infrastructure without worrying about a drop in response quality. With closed models, this option simply doesn't exist.

➡️ Model advantages:
• The 1M token context truly works. You can upload entire projects or batches of documents in one go, without splitting them into pieces.

• Chinese models historically cost 3-5 times less than Western ones, which provides significant savings for large query volumes.

• The ability to run on your own server also protects against sanctions — if Western services block Russian cards or legal entities tomorrow, Kimi will continue to work.

• The model natively understands Russian without a drop in reasoning quality.

Kimi itself is pleasant, with a clean interface. Although half a year ago there were bugs, and chat names were somehow labeled in Chinese 👉

You can test it here ⬅️

🌐 Website | 🎰LinkedIn