GLM 5.3 Flash pricing: API costs and cache
GLM-5.3-Flash costs $0.15 per million input tokens, $0.03 for cached input and $0.5 for output on Z.AI's direct API. A request with 100,000 uncached input tokens and 10,000 output tokens costs $0.02 before tools or other charges.
USD list prices checked 2026-11-04. These are token-cost examples, not measurements of model quality, speed or cost per completed task.
Choose the exact model before comparing prices
The official Z.AI table lists three separate products. All figures below are USD per million tokens.
| Model | Uncached input | Cached input | Output |
|---|---|---|---|
| GLM-5.3-Flash | $0.15 | $0.03 | $0.5 |
| GLM-5.3-FlashX | $0.37 | $0.075 | $1.25 |
| GLM-5.3 | $1.4 | $0.26 | $4.4 |
FlashX is a separately priced speed-oriented option. Its rates do not apply to ordinary Flash calls. GLM-5.3 is also a different model; shortening all three names to “GLM” makes a budget hard to audit. Match the model selected in your application with the name on your provider's invoice.
The GLM-5.3-Flash model guide describes a 1M-token context window. The context window is the capacity of one request. You pay for the input tokens you send, including large prompts.
What does 100K input plus 10K output cost?
Use this repeatable workload: 100,000 uncached input tokens and 10,000 billable output tokens per request, with no paid tools. Multiply each token category by its rate, then divide by one million. For Flash, input costs $0.015 and output costs $0.005, giving $0.020 per request.
| Same token workload | One request | 1,000 requests |
|---|---|---|
| GLM-5.3-Flash | $0.02 | $20 |
| GLM-5.3-FlashX | $0.0495 | $49.5 |
| GLM-5.3 | $0.184 | $184 |
These totals hold token counts constant. They do not show that one model can finish the same assignment for less money. A longer response, extra reasoning, retries or a failed attempt can change the amount of billable usage. Compare the total bill for an accepted result when evaluating your own workload.
Open Flash with these inputs in the calculator. It preselects the actual catalog model and 1,000 requests, yielding $20. Change requests to your monthly volume. The cache toggle assumes all input is an existing cache hit; leave it off for this uncached example.
Cached input is cheaper, but it is not free
The $0.03 cached-input rate is 80% below the $0.15 uncached-input rate. That percentage applies to the input tokens which receive cache-hit billing. It does not automatically reduce your complete invoice by 80%, because output retains its $0.50 rate.
For example, suppose an otherwise identical request records 80,000 cached input tokens, 20,000 fresh input tokens and 10,000 output tokens. Its token charge is $0.0024 + $0.003 + $0.005 = $0.0104. Across 1,000 such requests, that is $10.40. This assumes the cache hits actually occur; the initial request and cache misses need their own calculation.
If all 100,000 input tokens instead receive cached-input billing, the request costs $0.008, or $8 for 1,000 requests. That is a 60% reduction from the $20 uncached example. Use your provider's recorded cached-token usage rather than assuming every repeated prompt qualifies.
Z.AI labels cached-input storage “Limited-time Free.” That is separate from the paid cached-input token rate. It is also not a promise of permanently free storage. Recheck the table before forecasting a long-running deployment; this estimate does not invent a storage expiry date or a separate cache-write discount.
API usage, Coding Plan and older free Flash models
A Reddit pricing discussion asks for the cheapest way to use this model and discusses subscriptions and different providers. Those are different purchase routes. Community offers can help identify options, but their prices are not Z.AI's direct API list price.
Keep a Coding Plan subscription budget separate from pay-as-you-go API tokens. Compare the plan's supported tools, usage limits and current terms with your intended work. Do not convert a monthly subscription fee into an unlimited API allowance, or reuse a launch promotion after its stated period ends.
The same official price page lists older GLM-4.7-Flash and GLM-4.5-Flash as free. That does not make GLM-5.3-Flash free. Check the complete version string when a tutorial or screenshot advertises “GLM Flash” without a number.
Build a budget from your actual usage
Start with uncached input, recorded cached input, billable output and request count. Add paid tools separately: Z.AI currently lists built-in web search at $0.01 per use. Ten search uses would add $0.10, beyond the token estimate. Taxes and charges from another hosting provider are outside these examples.
The linked price table does not give this model a Batch discount or a separate long-context price band. Use the displayed token rates for this estimate. Check account-specific agreements before using a negotiated rate.
AI API Prices editors checked the linked official pricing and model documentation on October 12, 2026; no paid inference benchmark was run for this guide. Continue to the Zhipu GLM price table for other catalog models, or use the Flash calculator preset to adapt the example to your traffic.