Open models like GLM-5.3 do have practical benefits, but can they outweigh the risks?

Anthropic released a report stating that Z.ai's open-weight model GLM-5.3, released in August, can be freely downloaded and modified by anyone, posing a significant cyberattack risk. Supporters of open models argue that this helps prevent the concentration of power, allows researchers to observe the model's internal structure and chain of thought, and enables enterprises to process sensitive data locally, with costs also typically lower according to a Mozilla report. However, closed-source models are currently more capable; the top eight models on the Artificial Analysis LLM leaderboard are all closed-source, with open models lagging by only a few months.