Qwen3 API pricing per 1M tokens: 7-Max, Max, Flash
3 min readWritten by genmux API Team
Alibaba's Qwen3 family covers more ground than most teams realize — a flagship reasoner, a strong general model at a lower price, and a near-free flash tier. genmux exposes the live lineup through one chat.completions endpoint, so picking the right Qwen3 model is a model string change, not a new integration.
The lineup
| Model | Best for | Context | Thinking | Price per 1M tokens (in / out) |
|---|---|---|---|---|
qwen3-7-max | Heaviest reasoning, tool-free agents | 128,000 | yes | $2.25 / $6.75 |
qwen3-max | Strong general reasoning at lower cost | 131,072 | yes | $1.08 / $5.40 |
qwen3-6-flash | High-volume, low-latency, cheap | 131,072 | no | $0.045 / $0.36 |
All figures are what genmux bills, converted from credits at $0.0125/credit (the Scale pack rate). qwen3-7-max and qwen3-max beat Alibaba's own list price by a consistent margin that we can say so honestly; qwen3-6-flash is priced for access, not for undercutting a public number. The Plus and Coder variants (qwen3-plus, qwen3-coder-plus, qwen3-coder-turbo) are paused while the upstream vendor issue that took them down gets fixed — they'll come back on this page with real prices once they're serving traffic again.
Which one to pick
- Default to
qwen3-maxfor anything that needs real reasoning quality — it's the best balance of capability and price in the family, and it supportsenable_thinkingif a request needs step-by-step reasoning. - Reach for
qwen3-7-maxwhen a task needs the heaviest reasoning the family offers and cost per call matters less than getting it right. - Use
qwen3-6-flashfor classification, extraction, short replies, or any workload where you fire thousands of calls a day and latency matters more than depth.
Cheaper than Alibaba's own list price
Our pricing rule: we sit 10% under the model owner's own list price wherever our supply cost allows, and never below it. Checked against Alibaba Cloud Model Studio's public pricing, 2026-09-06:
| Model | Alibaba list price (in / out) | genmux price (in / out) | Saving |
|---|---|---|---|
| Qwen3.7-Max | $2.50 / $7.50 | $2.25 / $6.75 | 10% |
| Qwen3-Max | $1.20 / $6.00 | $1.08 / $5.40 | 10% |
qwen3-6-flash isn't in that table because its reference price is a proxy figure rather than a confirmed Alibaba list price — we don't make a saving claim for it.
Calling any of them the same way
from openai import OpenAI
client = OpenAI(base_url="https://api.genmux.tech/v1", api_key="nb_live_...")
resp = client.chat.completions.create(
model="qwen3-max",
messages=[
{"role": "system", "content": "You write minimal, idiomatic diffs."},
{"role": "user", "content": "Add input validation to this function:\n\n" + code},
],
max_tokens=800,
temperature=0.1,
)
print(resp.choices[0].message.content)
print(f"cost: {resp.usage.cost} credits ({resp.usage.cost_source})")Swap model for qwen3-7-max, qwen3-6-flash or any other id in the table above — messages, max_tokens, temperature, top_p, stop and stream all work identically across the family. enable_thinking only takes effect on the models marked "yes" above; sending it to qwen3-6-flash is silently ignored rather than erroring.
Curl, if you'd rather skip the SDK
curl https://api.genmux.tech/v1/chat/completions \
-H "Authorization: Bearer $GENMUX_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen3-6-flash",
"messages": [{"role": "user", "content": "Classify this review as positive, negative, or neutral: ..."}],
"max_tokens": 20
}'usage.cost on the response is the exact number of credits deducted; check it against the reservation genmux made before the call by inspecting GET /v1/generations/{id} afterward.
Getting started
Every Qwen3 model in the table above is live behind the same key as every other model on the platform, billed from one prepaid balance topped up today with USDT (card and SBP are being enabled). Model pages with live rates: Qwen3.7-Max, Qwen3-Max, Qwen3.6-Flash. Full endpoint reference is in the docs.