October 13, 2026
How GLM-5.3 picks its answer, shown in 3D: 78 layers light up with every read of the conversation
GLM-5.3, an open-weight model, has 743 billion parameters and 78 layers. On Oct 13, a Reddit user posted a 10-minute 3D video with voiceover: each layer is drawn as a tower, and a flash hits it when the model reads part of the conversation.
GLM-5.3, an open-weight model, has 743 billion parameters and 78 layers. On Oct 13, a Reddit user posted a 10-minute 3D video with voiceover: each layer is drawn as a tower, and a flash hits it when the model reads part of the conversation.
This is the first video in the series with a voiceover: the earlier ones had no narration.