GLM-4.6V-Flash

GLM-4.6V-Flash is the small member of the vision pair Z.ai published on 7 December 2025. Both models went out the same day, both carry an MIT licence, and both can be downloaded by anyone. The difference is size: the flagship GLM-4.6V holds 107.7 billion parameters, this one 10.3 billion. That difference decides who can actually use them, and the download counters show it. In a thirty-day window the small model was pulled about 82,600 times against roughly 4,000 for the flagship — twenty times more, even though neither costs anything to download. The reason is hardware, not price: 10 billion parameters fit on a single accelerator, and quantised community builds of this model run on a laptop, while 107 billion parameters need a server. Whichever model is technically better, only one of them is within reach of the people doing the downloading. In the API the same model is billed at zero. Z.ai keeps exactly three free text tiers — this one, GLM-4.5-Flash and GLM-4.7-Flash — and prices the paid vision line above them, with GLM-4.6V at $0.30 per million input tokens. The technical envelope is not a cut-down one. The context window is 131,072 tokens, the same order as the paid models, and the vision encoder is a full 24-layer stack at 336-pixel input with temporal patching, meaning the model reads video frames and not only stills. The text side is a dense 40-layer stack, 4,096 wide, on Z.ai's usual 151,552-token vocabulary.

▸News

Two models, same day, same licence: the small one was downloaded twenty times more

9/3/2026 · Hugging Face model repositories and Z.ai price list

▸Videos

No videos yet.

▸Reviews