What every GLM model costs, on Z.ai and every other host

It depends on the model. At Z.ai's own prices, the GLM models sold through OpenRouter run from $0.15 per 1M input tokens (GLM-5.3 Flash) to $1.40 (GLM-5.1). GLM-5.3 Flash is $0.15 / $0.50 per 1M input / output tokens on Z.ai, and the flagship GLM-5.3 is $1.40 / $4.40. Third-party hosts often charge less for the same model; the table above shows the cheapest host for each one, refreshed every 6 hours.

Through a third-party host. Right now the cheapest is $0.075 in / $0.25 out on DeepInfra (FP4), against Z.ai's own $0.15 / $0.50. Or self-host the MIT-licensed weights from Hugging Face (zai-org/GLM-5.3-Flash), which costs only your own GPU time.

Many GLM models have open weights (GLM-5.3 Flash, for one, is MIT-licensed), so any host can serve them and set its own price. Hosts compete on price, and some run lower-precision builds (FP4 instead of FP8) that are cheaper to serve and can answer slightly differently. Z.ai's own endpoint charges its published list price.

GLM-5.3 is the larger flagship at $1.40 / $4.40 per 1M input / output tokens on Z.ai. GLM-5.3 Flash is the smaller, faster model at $0.15 / $0.50, roughly 9x cheaper on input. Flash is the model that ran anonymously as Ox Alpha.

Z.ai's own API lists some older Flash models (GLM-4.7-Flash and GLM-4.5-Flash) as free, as of October 4, 2026. The current GLM-5.x models are paid on every host, but several publish open weights you can download and run yourself, including GLM-5.3 Flash under the MIT license.

Every 6 hours, from OpenRouter's public API, and each table shows when its prices were fetched. The Z.ai column is Z.ai's own endpoint on OpenRouter, which matches the list prices Z.ai publishes on its pricing page.