Qwen API price: a calculation for Qwen 3.8 Max and 3.8 Flash on typical tasks

When people search for "qwen api price", they usually do not want a line from a price list but an answer to "how much will this cost on my task". Below we calculate the cost of Qwen 3.8 Max and Qwen 3.8 Flash requests on three typical tasks with visible formulas. Prices are as of October 18, 2026 and change: current values are always in GET https://api.zukko.pro/v1/models or on the "Models" page in the dashboard.

Qwen prices on October 18, 2026

Prices per 1M tokens from the Zukko catalog and the provider's official prices from the same table:

The strength index is the Intelligence Index from Artificial Analysis. We give no other quality figures: whether a model suits your task, test on your own data.

Cost formula

cost = input × input price + output × output price
(divide all token counts by 1,000,000)

Cache enters the formula as a separate line: cache read × read price. But for Qwen in Zukko the cache read price equals the input price (0.088 for Max, 0.038 for Flash), so cache changes nothing on the bill. This differs from Claude, where a cache read is roughly ten times cheaper than input. So for Qwen there is no point building a scheme around cache to save money: you optimize the token volume instead. Cache in detail: key limits and cache↗.

Task 1. A chat or support reply

One request: 1,500 input tokens (instruction, history, question), 500 output tokens.

Qwen 3.8 Max (Zukko):

• input: 1,500 × 0.088 / 1,000,000 = $0.000132;
• output: 500 × 0.238 / 1,000,000 = $0.000119;
• total $0.000251.

Qwen 3.8 Flash (Zukko):

• input: 1,500 × 0.038 / 1,000,000 = $0.000057;
• output: 500 × 0.075 / 1,000,000 = $0.0000375;
• total $0.0000945.

At official prices: Max is 1,500 × 2 / 1,000,000 + 500 × 6 / 1,000,000 = $0.003 + $0.003 = $0.006; Flash is 1,500 × 0.15 / 1,000,000 + 500 × 0.47 / 1,000,000 = $0.000225 + $0.000235 = $0.00046.

If a chat makes 10,000 such replies a month, in Zukko that is 10,000 × $0.000251 = $2.51 on Max and 10,000 × $0.0000945 = $0.945 on Flash. For comparison, at official prices: $60 on Max and $4.60 on Flash.

Task 2. Bulk classification

100,000 texts, each about 400 input tokens (instruction plus text), with an answer of about 20 tokens (a label). Total 40M input and 2M output.

Qwen 3.8 Flash (Zukko):

• input: 40 × 0.038 = $1.52;
• output: 2 × 0.075 = $0.15;
• total $1.67.

Qwen 3.8 Max (Zukko):

• input: 40 × 0.088 = $3.52;
• output: 2 × 0.238 = $0.476;
• total $3.996, about $4.00.

At official prices: Flash is 40 × 0.15 + 2 × 0.47 = $6.00 + $0.94 = $6.94; Max is 40 × 2 + 2 × 6 = $80 + $12 = $92.

In money terms such classification costs a few dollars even on Max; whether Flash is good enough, check on a sample of your own texts. For comparison with similar cheap models: Qwen against GLM↗.

Task 3. Questions about a large document

A 200,000 token document and 10 questions about it. Each request sends the whole document (200,000 input tokens) and gets about 1,000 tokens back. Total 2M input and 10,000 output. This task needs only a text model, so both fit.

Qwen 3.8 Flash (Zukko): 2 × 0.038 + 0.01 × 0.075 = $0.076 + $0.00075 = $0.07675, about $0.077.

Qwen 3.8 Max (Zukko): 2 × 0.088 + 0.01 × 0.238 = $0.176 + $0.00238 = $0.17838, about $0.178.

At official prices: Flash is 2 × 0.15 + 0.01 × 0.47 = $0.30 + $0.0047 = $0.3047, about $0.305; Max is 2 × 2 + 0.01 × 6 = $4 + $0.06 = $4.06.

Note that the document is resent in every request. Cache will not lower the bill for Qwen in this scenario (the read price equals the input price), so savings come from elsewhere: ask several questions in one request, trim the document to the sections you need, or first search for relevant fragments. Ideas: RAG on a cheap model↗.

Summary table

The sums show orders of magnitude. Real spend depends on the length of your prompts and answers, so first measure tokens on several real requests: the usage field in the response, and before sending you can estimate with token counting↗.

Choosing between Max and Flash

The price gap between the models per the table: Max input is 2.3 times higher (0.088 versus 0.038), output 3.2 times (0.238 versus 0.075). Strength index 45 versus 40, and Max accepts images while Flash does not. If you need images, the choice is obvious: Max. If the input is text only, run a test: 50 to 100 of your examples on both models, compare on a metric you care about (label accuracy, share of valid JSON, a reviewer's score). If Flash passes the threshold, the whole price difference stays with you. If it does not, a cascade works: Flash first, with uncertain and hard cases sent to Max.

A first request to Qwen looks like an ordinary OpenAI SDK call:

import os
from openai import OpenAI

client = OpenAI(base_url="https://api.zukko.pro/v1", api_key=os.environ["ZUKKO_API_KEY"])
r = client.chat.completions.create(
    model="qwen3.8-flash",
    messages=[{"role": "user", "content": "Classify the review: \"Delivered fast, but the box was dented.\" Answer: positive, neutral or negative."}],
    max_tokens=10,
)
print(r.choices[0].message.content, r.usage.prompt_tokens, r.usage.completion_tokens)

FAQ

How much does the Qwen API cost per 1M tokens?

On October 18, 2026 in Zukko: Qwen 3.8 Max is $0.088 for input and $0.238 for output, Qwen 3.8 Flash is $0.038 for input and $0.075 for output. Official prices from our table: $2 / $6 and $0.15 / $0.47.

Does cache lower the Qwen price?

In Zukko the cache read price for Qwen equals the input price, so cache changes nothing on the bill. For Claude models it works differently.

How do I pay for the Qwen API?

Top up a dollar balance from $1: SBP (the payment page adds an 8.5% fee on top), MasterCard cards (fee about 13%, amounts $2 to $594) or crypto.

Where do I start?

Create a key in the dashboard, set a daily limit and repeat the calculation above on your own data.