Z.ai announced on August 28 that GLM-5.3 is now open-weight. The announcement came as a single post from the company's @Zai_org account.
That post is, so far, the whole story. No license terms, parameter count, context length or benchmark table accompanied it, and no independent coverage had landed at the time of writing — so treat any spec you see quoted elsewhere as unverified until Z.ai publishes a model card. The HN thread is where practitioners are comparing notes on what actually shipped.
What is worth noting without any numbers: the GLM line continues to arrive as downloadable weights rather than API-only, which keeps a frontier-adjacent Chinese model inside reach of anyone running their own inference. That is the part that matters for capacity planning, regardless of where 5.3 lands on the leaderboards.
Sources: Z.ai announcement, HN discussion