GLM-5.3-FlashX costs $0.37 in on Z.ai; Sume lists Flash, not FlashX

Short answer

If you are searching for GLM 5.3 Fast, the Z.ai pricing page read on 2026-10-08 does not show a row with that name. It lists GLM-5.3-Flash, GLM-5.3-FlashX and GLM-5.3. Sume's registry lists one GLM row, GLM 5.3 Flash, with the route z-ai/glm-5.3-flash. Sume does not list FlashX, the full GLM-5.3, or anything called Fast.

I cannot say whether Fast is a product name used elsewhere or a nickname for FlashX; the page I could read does not settle it.

Z.ai rows

All prices are per million tokens, as shown on the Z.ai page. Cached input storage is marked limited-time free there.

A turn on each

Using the Z.ai prices and a 30,000-input, 2,000-output turn: Flash is 30,000 x 0.15 / 1M = $0.0045 plus 2,000 x 0.50 / 1M = $0.0010, so $0.0055. FlashX is $0.0111 plus $0.0025, so $0.0136. GLM-5.3 is $0.042 plus $0.0088, so $0.0508. For comparison Claude Haiku 5.5 is $0.0040 on the same turn.

None of this is billed by Sume for FlashX or GLM-5.3, since you cannot select them there.

If you need them anyway

Call Z.ai from your own code and use Sume for the media. The Agent Completions model field accepts only sume-agent, so the integration point is an HTTP call per task, not a model switch.

Why a cheap tier still needs a plan

Flash-class models are tempting for an agent that mostly routes and formats. Z.ai's page puts Flash at $0.15 per million input tokens, which is 1.5 times Haiku 5.5's $0.10, and at the same $0.50 output. So on Z.ai's own prices Flash is not cheaper than Haiku 5.5 at the base tier; it is a different model with different strengths, and I have no benchmark to compare them.

The same arithmetic applies on Sume's side if you select the GLM row, except that Sume's card is lower. See the card post linked below for the exact gap.

To keep this grounded, here is what each source supports. Z.ai's pricing page, read 2026-10-08, supports the rows it prints, including Flash and FlashX. Sume's registry supports one GLM row, GLM 5.3 Flash, with its own rate card. Nothing I read supports a row named Fast on either side, so I do not write about one. If you saw GLM 5.3 Fast in a launch post, check whether the post means FlashX or Flash before you plan a budget around it, because their prices differ on the Z.ai page and only Flash is selectable in Sume.

In short, treat any row you cannot find on a vendor page or in Sume's picker as unverified, and say so in your own notes before it reaches a budget.

Re-read both pages when you plan, since vendor prices move and this post is a snapshot from 2026-10-08.

• Haiku 5.5 base input $0.10 vs GLM-5.3-Flash on Z.ai $0.15.
• Output equal at $0.50 per million.