Moonshot AI · Model release
Kimi K3: open weights, huge context, and a datacentre-sized trade-off
Kimi K3 brings frontier ambitions, multimodal input, and open weights — but its scale changes what self-hosting means.
What changed
- The model launched as a long-context flagship and later gained downloadable weights.
- Its Mixture-of-Experts design activates fewer parameters than its total size suggests, but the full checkpoint is still enormous.
- The Kimi licence is not the same thing as MIT or Apache licensing.
Best fit
- Long repository analysis
- Open-weight evaluation
- Multimodal agent work
What to be careful about
- Self-hosting is a multi-node decision, not a normal workstation install.
- Review the Kimi K3 licence before commercial redistribution or hosted access.
Dataset snapshot: Kimi K3 is also represented in the checked comparison dataset. Prices and benchmark figures carry their own field dates.