Answer: B
China's increasing use of open-weight LLMs, including models such as Qwen, DeepSeek and Kimi.
During training of the AI model, it processes enormous amounts of data and adjusts billions of numerical values called weights.
These weights contain what the model has learned about patterns in the training data.
A simplified picture:
Training data
↓
Training process
↓
Billions of parameters/weights
↓
Trained AI model
So, the weights are essentially the learned numerical parameters of the model.
For example, Company A keeps the trained weights private.
You may be allowed to use the model through an API:
You send a question → Company's server runs the model → You receive the answer.
You don't receive the actual trained model weights. We can assume the operation of ChatGPT and other LLMs.
In the Open-weight model, the Company A makes the trained weights available.
You may be able to:
Download weights → Run the model yourself → Modify/fine-tune it → Deploy it on your own infrastructure.
For example:
A general language model can be fine-tuned into a Tamil legal-document model with specific inputs.
Also, the exact permissions depend on the licence.
Some of the open weight AI models include the GLM/Kimi/DeepSeek. China is leading in this field.
Sarvam AI — India's leading open-weight AI lab, based in Bengaluru. It was founded by Dr. Vivek Raghavan and Dr. Pratyush Kumar.
September 17, 2026 1.2K 6
