Z.ai

Z.ai published weights for fourteen of its sixteen rows in the OpenRouter catalog, as observed on 31 August 2026. The two it did not are both called Turbo, and a checkpoint with no published weights has no third party able to host it — so on those two rows the listed price is Z.ai's own, and on the open rows it need not be. z-ai/glm-5, listed 11 February 2026, carries a Hugging Face id and heads at $0.60per million tokensopenrouter-models, last checked 2026-09-14 for input; z-ai/glm-5-turbo, listed a month later on 15 March, carries no weights and lists at $1.20per million tokensopenrouter-models, last checked 2026-09-14. z-ai/glm-5v-turbo, the multimodal one listed 1 April, matches it to the digit. The open row's figure is the top listed provider's rate rather than necessarily Z.ai's own, so the distance between the two is not a measure of what this company charges to keep a checkpoint to itself: it is a closed price quoted against whichever host happens to head an open row.

The other field Z.ai fills in that almost nobody does is the death date. Of the 396 rows in the catalog, as observed on 31 August 2026, six carried a non-null expiration date and five of them were Z.ai's; no other vendor carried more than one. It uses the field two ways. z-ai/glm-4.5, one half of the July and August 2025 pair that carried the company's reputation abroad, reads 2026-12-31openrouter-models, last checked 2026-09-14, a real retirement seventeen months after listing; its sibling z-ai/glm-4.5v no longer carries a non-null expiration date in this snapshot, so the pair no longer agrees. The four current rows, both Turbos plus z-ai/glm-5.3 and z-ai/glm-5.3-flash, read not publishedopenrouter-models, last checked 2026-09-14: a sentinel seventy-two years out, meaning nothing is planned. Neither date is printed on the listing page a buyer reads. Both sit in the API row behind it.

The cheap row is the capable one. z-ai/glm-5.3, listed 18 August 2026, is text-only and lists at $1.40per million tokensopenrouter-models, last checked 2026-09-14 for input. z-ai/glm-5.3-flash, listed eight days later, takes text, images and video, advertises the identical window of 1310720tokensopenrouter-models, last checked 2026-09-14, and lists at $0.15per million tokensopenrouter-models, last checked 2026-09-14 — around a nineteenth as much. Z.ai credits a "hybrid sparse and linear attention architecture" for holding long-context behaviour down there. Within one week the company shipped a model that reads more kinds of input, over the same window, on the cheaper row.