Model Licenses
Last Updated: September 28, 2026
The models available through Kurrens are open-weight models developed by third parties and released under the licenses listed below. By using a model through the Kurrens Services, you agree to comply with that model’s license and any acceptable use policy that accompanies it, as described in Section 5 of our Terms of Service. Model developers do not endorse Kurrens.
Where a model’s license or use policy is stricter than our Acceptable Use Policy, the stricter rule applies.
Keep this list in sync with the live catalog (
site/src/data/models.catalog.json). Rows marked for review must be cleared by counsel before the model goes live.
MIT License
| Model | Developer | License |
|---|---|---|
| DeepSeek-V4.1-Flash, DeepSeek-V4-Pro-0813, DeepSeek-V4-Flash-0731 | DeepSeek | MIT |
| GLM-5.3-Flash, GLM-5.2 | Zhipu AI (Z.ai) | MIT |
| MiMo-V2.6-Pro, MiMo-V2.6-Flash | Xiaomi | MIT |
Apache License 2.0
| Model | Developer | License |
|---|---|---|
| Qwen3.8-27B, Qwen3.6-35B-A3B, Qwen3-Coder-Next | Alibaba Qwen | Apache-2.0 |
| gpt-oss-120b, gpt-oss-20b | OpenAI | Apache-2.0 · usage policy |
| Gemma 4 31B IT | Apache-2.0 |
Other Licenses (counsel review required before launch)
| Model | Developer | License | Notes for review |
|---|---|---|---|
| GLM-5.3 | Zhipu AI (Z.ai) | GLM-5.3 License | Custom license (glm-5.3); full text not yet reviewed |
| Kimi-K3 | Moonshot AI | Kimi K3 License | Full text not yet reviewed |
| Kimi-K2.7-Code, Kimi-K2.6 | Moonshot AI | Modified MIT | Attribution requirement above usage/revenue thresholds |
| MiniMax-M3 | MiniMax | MiniMax Community License | Custom license; full text not yet reviewed |
| MiniMax-M2.7 | MiniMax | License | Custom license; full text not yet reviewed |
| Qwen3.8-Flash-Next | Alibaba Qwen | Qwen Community License 1.0 | Not Apache-2.0, unlike earlier Qwen releases; check commercial-use thresholds |
| Qwen3.8-2.4T-A95B | Alibaba Qwen | Qwen3.8 Max License | Custom license; full text not yet reviewed |
Quantization Notice
Some models are served at reduced numerical precision (for example, FP8). The precision for each model is shown on its model page and in our API model listing. Quantized weights are derived from the original weights published by the developer and remain subject to the original license.