# Open Weights

> An open-weights model is one whose trained parameters (the weights) are published for download, so you can run and usually fine-tune them, subject to the license. It is not the same as open source: the weights are released, but the training code, data, and recipe usually are not, so you cannot fully reproduce or audit how it was built.

- Updated: 2026-07-28
- Canonical: https://geotoolbox.ai/glossary/open-weights

---

"Open weights, not open source" is the distinction most coverage gets wrong. Models like DeepSeek, Qwen, and Kimi K2 release downloadable weights, often under permissive licenses, but that is a different thing from the open-source software definition, which also expects the source and the ability to rebuild from scratch.

Why it matters: open weights let a company self-host a capable model, fine-tune it on private data, and run it without sending anything to a vendor. The license still governs commercial use, and some are "modified" (Kimi's K2-family license adds an attribution requirement at scale, and the custom Kimi K3 License that shipped with its 2026 flagship goes further, requiring a separate agreement once a model-as-a-service operator passes $20 million in revenue over any 12 months). Always read the model card rather than trusting the "open source" label, which is applied loosely across the industry.