Z.AI API pricing

Z.AI (formerly Zhipu AI) is one of the few providers where most of the current lineup is genuinely open-weight — the GLM line ships its weights publicly under a permissive license, with a smaller set of "turbo" variants held back as closed, API-only tiers. See the live table above for current rates, and the open-vs-closed race to see exactly how competitive Z.AI's open releases are against closed frontier models.

Open by default, "turbo" is the exception

The numbered GLM releases (GLM-4.5, GLM-4.6, GLM-4.7, GLM-5, GLM-5.1, GLM-5.2 and their variants) are open-weight — downloadable and self-hostable, which is unusual among providers competing at the frontier. The "-turbo" suffixed variants (e.g. a GLM-5-turbo or GLM-5v-turbo) are the closed, API-only counterparts, typically optimized for latency or cost rather than being materially different models. If self-hosting or license flexibility matters to you, confirm you're looking at a non-turbo release before assuming it's open.

Why this matters for pricing

Because most of the lineup is open-weight, Z.AI's own API pricing tends to be aggressive — it's competing against the possibility that a customer simply self-hosts the same weights elsewhere. This is a useful signal in general: providers whose flagship models are also freely downloadable usually can't sustain a large API markup, since price-sensitive customers have a credible alternative.

Practical guidance

• Default to the latest numbered GLM release for general chat and coding — the line has been competitive on both LiveBench and coding-specific benchmarks like SWE-bench Verified.
• Check vision/tool-calling variants separately (GLM releases with a "V" designation add multimodal input) if your workload needs image understanding — these are typically separate SKUs from the text-only line.
• Don't assume "turbo" means better — it usually means a closed, latency-optimized variant, not a capability upgrade.

Rate limits

Z.AI tiers API access similarly to its peers, with caps that scale by account tier and plan. We don't have independently-verified specifics for Z.AI's exact current tier structure, so confirm requests-per-minute and token caps directly in the Z.AI console before capacity-planning a production workload — this is generally good practice across every provider, since tiers shift as capacity and demand change.

See how this compares to the other 8 providers we track on the rate limits page.

When Z.AI vs. a competitor

For coding and general-purpose work, compare GLM directly against DeepSeek, Qwen, and Moonshot — the cluster of providers where open-weight, competitively-priced models are the norm rather than the exception. See how GLM's coding benchmarks stack up on SWE-bench Verified and Aider Polyglot, or put it head-to-head with any model on compare.

See current GLM pricing live above, or find where it ranks on cheapest models.