Z.ai lists GLM-4.6 at $0.60 per million input tokens and $2.20 per million output tokens, in USD. Its pricing table separately lists cached input at $0.11/M tokens and cached-input storage as “Limited-time Free.” The page does not specify the promotion’s end date or eligibility terms. These are token-based API rates—not one universal price for every service that hosts or routes the model. (docs.z.ai)
For a budget, choose the provider and API route first, then apply its rates to expected input, cached-input, and output tokens. Third-party services may list different prices and terms. Dedicated GPU hosting is a separate cost model; comparing it with token pricing requires workload and utilization assumptions.
GLM-4.6 API prices by route
| Route or listing | Input | Cached input | Cached-input storage | Output | What the price represents |
|---|---|---|---|---|---|
Z.ai API |
$0.60/M tokens | $0.11/M tokens | Limited-time free; duration and eligibility terms aren’t specified in the captured pricing information | $2.20/M tokens | Z.ai’s published model rates, in USD per million tokens. (docs.z.ai) |
| OpenRouter listing | $0.50/M tokens | $0.10/M tokens | Not specified here | $2.00/M tokens | OpenRouter’s provider table labels the $0.10/M DeepInfra rate as cache input; OpenRouter also lists other providers at different rates. (openrouter.ai) |
| Other aggregator listings | Varies | Varies | Varies | Varies | Treat these as route-specific listings, not as Z.ai’s official price; check the chosen service’s terms before budgeting. |
Z.ai’s pricing table distinguishes Cached Input from Cached Input Storage. The $0.11/M figure is the cached-input token rate; “Limited-time Free” refers to cached-input storage, not to cached input being free. The source does not provide the promotion’s end date or full terms. (docs.z.ai)
OpenRouter likewise shows why the route matters: its table lists DeepInfra at $0.50/M input, $0.10/M cached input, and $2.00/M output, while listing other providers at different rates. Those are OpenRouter provider listings, not a replacement for Z.ai’s published prices. (openrouter.ai)
A captured AIHubMix listing also displays provider- and tier-specific rates, including a Z.AI row whose output threshold is shown as “200” without a unit. A captured NinjaChat listing gives $0.50/M input, $0.05/M cached input, and $2.00/M output. Treat these as service-specific listings rather than current direct-provider quotes; confirm the chosen route’s price and tier before relying on them.
Estimate GLM-4.6 API spend
Use the selected provider’s rates and count uncached and eligible cached input separately:
(uncached input tokens ÷ 1,000,000 × input rate)
+ (eligible cached input tokens ÷ 1,000,000 × cached-input rate)
+ (output tokens ÷ 1,000,000 × output rate)
Only apply the cached-input rate to tokens the provider confirms qualify. Do not count the same input tokens as both uncached and cached. Confirm how cache eligibility and billing are determined; the Z.ai pricing information also labels cached-input storage “Limited-time Free” but does not state the promotion’s duration or full terms.
Illustrative arithmetic using Z.ai’s listed rates: for 1 million cached input tokens and 200,000 output tokens, the calculation is $0.11 + $0.44 = $0.55. This is an estimate from listed rates, not a provider quote. It assumes the input qualifies for cached-input pricing and excludes any other charges. The storage promotion is separate, and its duration and eligibility terms are not specified in the captured pricing information. (docs.z.ai)
For a monthly estimate, apply the formula to your expected monthly request mix—not just the number of requests. Long system prompts, conversation history, tool interactions, and generated answers can change token totals. Check the chosen provider’s terms for account-specific discounts, taxes, and request options that may affect charges.
Token pricing versus dedicated GPU hosting
Dedicated hosting is a different purchase. FlexAI’s GLM-4.6 listing describes a dedicated 16× H100 deployment rather than per-token billing. The captured page body and its search-result snippet show conflicting hourly rates, so neither should be treated as a definitive current quote without confirmation from FlexAI. Even a confirmed hourly price would not be directly comparable with API rates without workload, utilization, scaling, and any minimum or additional-charge assumptions.
The FlexAI listing also says that its described deployment is not served through the shared Token Factory API. That statement does not establish whether another Token Factory route exists or what it would cost.
Production API: what the available information establishes
Z.ai’s model documentation describes GLM-4.6 as having a 200K-token context window and a maximum output of 128K tokens. Those are model-page specifications, not proof that every third-party host supports the same limits or features. Check the documentation for your chosen route if your application depends on a particular context size, tool behavior, or output limit.
Z.ai’s general quick start documents API keys, REST calls, SDK options, and an OpenAI-compatible SDK path. Its example request uses glm-5.3, not glm-4.6; that example alone therefore does not confirm the GLM-4.6 model identifier or availability for your account. Confirm the model name and endpoint in current documentation before integrating.
Third-party examples apply to those services only. The captured AIHubMix Python example uses AIHubMix’s own base URL and API-key variable; the captured NinjaChat cURL example uses NinjaChat’s endpoint and Bearer key. Neither is direct Z.ai authentication guidance or proof of production readiness.
Before committing a production budget, confirm with the provider you intend to use:
- Availability and identifier: GLM-4.6 is enabled for your account, and you have the correct model name and endpoint.
- Applicable rates: input, output, cached-input pricing, tier boundaries, and whether rates differ by route or region.
- Billing details: cache eligibility, cached-input storage promotion terms, minimums, rounding, taxes, and request options that affect charges.
- Operational fit: supported context and output limits, required capabilities, rate limits, service terms, and regional availability.
- Observed workload: token usage and latency on representative requests through the intended production route.
Bottom line: Z.ai lists GLM-4.6 at $0.60/M input and $2.20/M output, with $0.11/M cached input and cached-input storage marked “Limited-time Free.” The promotion’s duration and eligibility terms are not specified in the captured pricing information. For a third-party API, use that route’s own listed price and terms. Treat conflicting dedicated-compute figures and unclear aggregator tiers as items to confirm—not as a universal GLM-4.6 price.
Sources
Z.AI Developer Documentation — Pricing
OpenRouter — GLM 4.6 API Pricing and Providers
Z.AI Developer Documentation — GLM-4.6 Overview
Z.AI Developer Documentation — Quick Start