GLM-5.2 drops open weights under MIT

The launch of GLM-5.2 demonstrates that the gap between open-weight and proprietary models is closing, shifting the battleground from raw benchmark scores to data privacy and deployment autonomy.

* Releasing under an MIT license grants developer communities full ownership and customization capabilities over their agent infrastructure.

* The IndexShare architecture represents a significant optimization for executing long-context operations without proportional increases in hardware costs.

* Incorporating variable thinking levels shows a developer-first focus on managing inference cost and latency depending on task difficulty.