China AI Hub China AI Hub

Models / Open-weight Chinese AI models

Open-weight Chinese AI models

Open-weight models publish their trained weights for download and self-hosting, unlike API-only models. The database records open-weight status and license terms per model so builders can evaluate deployment rights against the actual license — MIT, Apache-2.0, or a custom regime.

15
models match this dimension
Counted automatically from the models database: open_weight === true

The license, not the weights alone, decides what you can legally do: MIT and Apache-2.0 are permissive, while the custom regimes (Kimi K3, MiniMax M-series, Qwen3.8) attach revenue or attribution thresholds.

Editorial position

Our read: the licence, not the weights, is the decision. Apache-2.0 and MIT models are deployable without a conversation; the custom Chinese licences — MiniMax's community terms, Kimi's revenue-triggered clauses, several Qwen variants — turn self-hosting into a legal review. Chinese labs use open weights as distribution, not as a gift, so expect permissive licences on the models they want adopted and restrictive ones on the models they want to monetise.

Pattern in the data

Distribution of this dimension across matching models
Models matching15
Open weights15 of 15
Providers represented6
Context window range200K – 1024K
Input price range /1M$0.15 – $26
Licence familiesMIT, Apache-2.0, Modified, Kimi, Custom, MiniMax, Qwen3.8-Max

The data shows how completely open weights have become the default in China: most tracked models publish weights, and the licences split cleanly into permissive (MIT, Apache-2.0) and custom regimes with commercial conditions. The price range among open models is wider than among closed ones, which is the signature of labs using weights as distribution rather than as a product.

Matching models

GLM-5.2

zhipu-ai · deprecated

  • Context: 1,048,576 tokens
  • Open weight · MIT (pure open, no regional limits per official model card)

GLM-5.3

zhipu-ai · active

  • Context: 1,048,576 tokens
  • Open weight · Apache-2.0 (per GitHub repo metadata; README has no separate weights-license section - verify per-model HF cards before reuse)

GLM-5.3-Flash

zhipu-ai · active

  • Context: 1,048,576 tokens
  • Open weight · Apache-2.0 (per GitHub repo metadata; README has no separate weights-license section - verify per-model HF cards before reuse)

Kimi K3

moonshot-ai · active

  • Context: 1,048,576 tokens
  • Open weight · Kimi K3 License (permissive MIT-style, but Model-as-a-Service operators with >$20M aggregate revenue over any 12 months must sign a separate agreement; products with >100M MAU or >$20M monthly revenue must display 'Kimi K3' in the UI)

MiniMax-M2.7

minimax · active

  • Context: 204,800 tokens
  • Open weight · Custom NON-COMMERCIAL license (MIT-style terms for non-commercial use only; any commercial use requires prior written authorization from MiniMax at [email protected]; attribution 'Built with MiniMax M2.7' required)

MiniMax-M2.7-Highspeed

minimax · active

  • Context: 204,800 tokens
  • Open weight · Custom NON-COMMERCIAL license (MIT-style terms for non-commercial use only; any commercial use requires prior written authorization from MiniMax at [email protected]; attribution 'Built with MiniMax M2.7' required)

MiniMax-M3

minimax · active

  • Context: 1,048,576 tokens
  • Open weight · MiniMax Community License (custom): free for non-commercial use; commercial use requires prominent attribution 'Built with MiniMax M3' plus written authorization from MiniMax if yearly revenue exceeds US$20M (otherwise a one-time notice to [email protected])

Qwen3.8-2.4T-A95B

alibaba-cloud · active

  • Context: 262,144 tokens
  • Open weight · Qwen3.8-Max License (custom MIT-style: unrestricted use/copy/modify/sell, but products with >100M MAU or >US$20M/month revenue must display the model name; Model-as-a-Service or AI Work Assistant businesses with >US$50M/12-month revenue need a separate license from Qwen)

Decision guides

Original research

This page is generated from the models database; the member list and count are derived from each model's frontmatter fields, not written by hand. See the full model database and the methodology.