GLM API price: what GLM 5.3 and 5.3 Flash cost on real tasks

If you are looking for the GLM API price, meaning what requests to GLM 5.3 and its cheaper version 5.3 Flash cost, the answer is short but with nuances. In Zukko these models have a single price for input and output, and cache is counted at the same price. Below are the October 18, 2026 price table, the formula and three scenarios with arithmetic: chat, long generation and batch processing.

GLM prices in Zukko

Prices in $ per 1M tokens:

Official provider prices for comparison: glm-5.3 is $1.4 for input and $4.4 for output, glm-5.3-flash is $0.15 and $0.5. Provider prices change, so this is an example for the stated date. The current list is GET https://api.zukko.pro/v1/models or the "Models" page in the dashboard.

Two features that matter for the calculation:

• Input and output cost the same. In many catalogue models output is roughly 2–5 times more expensive than input, and long answers noticeably inflate the bill. GLM has no such skew: you pay for the total number of tokens.
• Cache gives no discount. The cache read price equals the input price. A read is counted if the provider reports it in the response, but you cannot save on repeated context with GLM. How the gateway accounts for cache across model families is described on the key limits and cache↗ page.

The formula

For GLM it collapses to one line:

cost = (input tokens + output tokens) × price / 1,000,000

where the price is $0.1 for glm-5.3 and $0.04 for glm-5.3-flash. Take the token count from the usage field of the response or estimate in advance via POST /v1/messages/count_tokens (its capabilities are described in the Zukko documentation). Russian text usually has more tokens than words, so leave a margin.

Scenario 1. A chat answer

A question with history: 1,000 input tokens and 1,500 output tokens, 2,500 in total.

• glm-5.3: 2,500 × 0.1 / 1,000,000 = $0.00025
• glm-5.3-flash: 2,500 × 0.04 / 1,000,000 = $0.0001

For $1 you get about 4,000 such answers on glm-5.3 (1 / 0.00025 = 4,000) and about 10,000 on glm-5.3-flash (1 / 0.0001 = 10,000). This is price arithmetic, not a quality promise: whether Flash is enough for your task, check on examples.

Scenario 2. Long generation

An article or report: 3,000 input tokens and 12,000 output tokens, 15,000 in total.

• glm-5.3: 15,000 × 0.1 / 1,000,000 = $0.0015
• glm-5.3-flash: 15,000 × 0.04 / 1,000,000 = $0.0006

For comparison, take claude-sonnet-5-5 ($0.4 input, $2 output) on the same task without cache:

• Input: 3,000 × 0.4 / 1,000,000 = $0.0012
• Output: 12,000 × 2 / 1,000,000 = $0.024
• Total: $0.0252

The effect of a single price is clearly visible here: for GLM a long output barely changes the bill, whereas for a model with expensive output it makes up the bulk. But these models differ in context (128K for GLM, 1M for Sonnet 5.5) and Intelligence Index (45 and 42 for GLM against 48 for Sonnet 5.5). Price is not the only criterion.

Scenario 3. A batch of 100,000 texts

Classification or a short summary: 100,000 documents, 1,500 input tokens and 300 output tokens each.

• Input tokens: 100,000 × 1,500 = 150,000,000 (150M)
• Output tokens: 100,000 × 300 = 30,000,000 (30M)
• Total: 180M tokens

Cost in Zukko:

• glm-5.3: 180 × 0.1 = $18
• glm-5.3-flash: 180 × 0.04 = $7.20

At the provider's official prices:

• glm-5.3: 150 × 1.4 + 30 × 4.4 = $210 + $132 = $342
• glm-5.3-flash: 150 × 0.15 + 30 × 0.5 = $22.5 + $15 = $37.5

If each document's context fits within 128K (and in this scenario it is far smaller), the batch runs without splitting. How to evaluate bulk tasks is shown in detail in the cheapest models for bulk tasks↗.

Which GLM to take

Choosing between glm-5.3 and glm-5.3-flash is a matter of a test, not a guess:

1. Take 50 of your own examples.
2. Run them on glm-5.3-flash and measure the share of correct answers.
3. If that is not enough, run them on glm-5.3 and calculate whether the quality difference pays for the $0.06 per million tokens difference.

If you need image input or a context above 128K, GLM does not fit: neither model accepts images, and context is capped at 128K. Look at other families then. A comparison with Chinese competitors is in Qwen versus GLM↗.

How to send a request

GLM is called like any model in the catalogue. In the OpenAI format:

from openai import OpenAI

client = OpenAI(base_url="https://api.zukko.pro/v1", api_key="fetch_your_key")

response = client.chat.completions.create(
    model="glm-5.3-flash",
    messages=[{"role": "user", "content": "Briefly describe what prompt caching is"}],
)

u = response.usage
total = u.prompt_tokens + u.completion_tokens
print(response.choices[0].message.content)
print("tokens:", total, "cost, $:", total * 0.04 / 1_000_000)

The last line computes the cost at the Flash price on October 18, 2026. If the price changes, substitute the current one.

Pay as you go: how payment works

There is no subscription; you pay for tokens from a dollar balance. You can top up from $1:

• SBP: rubles at the market rate, the payment page adds an 8.5% fee on top; the dollar amount you requested is credited to the balance. For example, a $10 top-up via SBP costs $10 × 1.085 = $10.85 in ruble equivalent;
• cryptocurrency (USDT and others);
• MasterCard cards: a fee of about 13%, amounts from $2 to $594.

Access from Russia goes through the gateway. Create a key on the keys page↗. Each key can have a daily and a monthly limit and a list of allowed models, which is convenient for batch jobs. More on models and prices: models documentation↗.

FAQ

Why do GLM input and output cost the same?

That is how the price list is built: glm-5.3 has both prices at $0.1, glm-5.3-flash at $0.04. The practical conclusion: long answers do not inflate the bill more than long requests do.

Does cache work with GLM?

The cache read price equals the input price, so there is no discount. If saving on repeated context matters to you, compare other models using the formula with cache.

What is the official GLM 5.3 price?

As of October 18, 2026: $1.4 for input and $4.4 for output per 1M tokens. Check the provider's current tariff, as it may change.

Can I send images to GLM?

No, neither model supports image input. For images choose models marked "image input: yes" in the model list.