Product descriptions in bulk
Write a product description under 60 words and five benefit bullets strictly from the specs: <specs>
system: You write for an online store: specific, no superlatives.max_tokens: 400
Alibaba · Text · API
Qwen API — live pricing and a 5-minute setup
Qwen API (Alibaba): Qwen 3.8 Max and Flash through one Zerocoder key. OpenAI-compatible chat/completions, streaming, function calling.
Qwen is Alibaba's family of language models: the Max flagship for hard tasks, code and long documents, and the fast, light Flash for high-volume scenarios. Through Zerocoder both are served on the OpenAI-compatible /v1/chat/completions with streaming and function calling — no Alibaba Cloud account needed.
You pay per token — input and output at their own prices — with a minimum per call, and a response that never starts is refunded. Below are the live prices of both models, a cost calculator and code samples.
Models and prices
The live GET /v1/models price list in US dollars. Minimum per call: $0.00556. Endpoints: /v1/messages, /v1/chat/completions
| Model | Price | Unit |
|---|---|---|
Qwen 3.8 Maxrecommended qwen3.8-max · Alibaba | input $2.04 / output $6.12 | per 1M tokens |
Qwen 3.8 Flash qwen3.8-flash · Alibaba | input $0.153 / output $0.479 | per 1M tokens |
All prices are in US dollars; Zerocoder prices are converted at the internal rate.
Cost calculator
Pick a model and a volume — the calculator uses the same formulas as API billing.
per call$0.00714
per month · 1,000$7.14
input $2.04 / output $6.12 per 1M tokens
minimum per call: $0.00556
Prices from GET /v1/models in US dollars at the internal rate. Failed generations are not charged.
- Get a keyin the API dashboard
- Top up the walletby card or invoice
- Make the callPOST /v1
/chat /completions, model qwen3.8‑max
First request
Connect the Qwen API: change two values
The OpenAI SDK in Python and Node.js, the Anthropic SDK or curl: swap the base URL and the key — requests, streaming and response handling stay as they are. Qwen answers on POST /v1/messages and /v1/chat/completions.
Base URLhttps://zerocoder.comAnthropic SDK — https://zerocoder.com, without /v1
Model in the samples — qwen3.8-max; any model from the table works
from openai import OpenAI client = OpenAI( base_url="https://zerocoder.com/api/v1", api_key="zc-sk-…",) r = client.chat.completions.create( model="qwen3.8-max", messages=[{"role": "user", "content": "Hello!"}],)print(r.choices[0].message.content)import OpenAI from "openai"; const client = new OpenAI({ baseURL: "https://zerocoder.com/api/v1", apiKey: "zc-sk-…",}); const r = await client.chat.completions.create({ model: "qwen3.8-max", messages: [{ role: "user", content: "Hello!" }],});console.log(r.choices[0].message.content);import Anthropic from "@anthropic-ai/sdk"; const client = new Anthropic({ // without /v1: the SDK appends /v1/messages itself baseURL: "https://zerocoder.com/api", apiKey: "zc-sk-…",}); const msg = await client.messages.create({ model: "qwen3.8-max", max_tokens: 300, messages: [{ role: "user", content: "Hello!" }],});console.log(msg.content[0].text);# Anthropic Messages formatcurl https://zerocoder.com/api/v1/messages \ -H "x-api-key: zc-sk-…" \ -H "content-type: application/json" \ -d '{"model": "qwen3.8-max", "max_tokens": 400, "messages": [{"role": "user", "content": "Hello!"}]}' # OpenAI chat/completions format, streamingcurl -N https://zerocoder.com/api/v1/chat/completions \ -H "authorization: Bearer zc-sk-…" \ -H "content-type: application/json" \ -d '{"model": "qwen3.8-max", "stream": true, "messages": [{"role": "user", "content": "Hello!"}]}'
How to connect the Qwen API: first request in 5 minutes
From a key to production — six steps based on the real /api/v1 contracts. Each step links to the relevant docs section.
- 1
Get a key in your account
Sign up, open the API section of your account, top up the balance and create a zc-sk-… key. One key unlocks Qwen and every other model in the catalogue.
Read more → - 2
Send the first request
POST /v1/chat/completions (Bearer) or /v1/messages (x-api-key) with model and messages — OpenAI and Anthropic shapes, the SDKs work with a swapped base URL.
Read more → - 3
Turn on streaming
stream: true — the answer arrives as SSE chunks with a final billing event showing the real charge. The reserve is taken by max_tokens (default 8192), so set it to fit the task.
Read more → - 4
Handle limits and errors
30 requests per minute per key: on 429 wait and retry with a growing pause; 402 — top up the balance; 400 — check model against GET /v1/models. Failed generations are not charged, so retrying is safe.
Read more → - 5
Watch the budget
GET /v1/balance returns the balance and 30-day spend; every call's price comes back in the response — write it to your logs and set an alert threshold in your own monitoring.
Read more → - 6
Ship to production
Keep the key in server secrets, never in the frontend. For high Qwen volumes contact us — the limit is raised individually and companies can pay by invoice. Check the observed stability on the status page.
Read more →
No code
Any tool that can make an HTTP request can call our API: one POST with the x-api-key header.HTTP Requestn8n
HTTP Request node: POST to the /api/v1 endpoint, x-api-key header and a JSON body built from workflow fields. For video add a Wait node and a second HTTP Request on GET ?id= in a loop until completed.
HTTP → Make a requestMake
HTTP “Make a request” module: POST, Body type JSON, x-api-key header. The response is parsed automatically — the image url or the answer text flows into the next modules.
Webhooks by ZapierZapier
Webhooks by Zapier → Custom Request action: POST, Data — JSON with model and prompt, Headers — x-api-key. Response fields are available to the following Zap steps.
Apps ScriptGoogle Sheets
In Apps Script use UrlFetchApp.fetch with method post, contentType application/json and a payload from the cells: prompts in one column, results written to the next one.
any bot frameworkTelegram bot
A bot on aiogram, grammY or Telegraf calls the same HTTP API: user message → model request → reply in the chat. For video — poll the status and send the file from the finished URL.
OpenAI-compatible base URLPython / Node SDK
For text models the official OpenAI and Anthropic SDKs work with a swapped base URL and our key — your application code stays the same. Images and video are a plain HTTP call from any language.
For business
We'll help bring AI into your business
More than a key: we help wire Qwen and other models into your workflows — from audit to production, with invoice payment.
- Process auditWe find where AI pays off in your workflows: support, sales, content, documents — and name specific scenarios and models.
- Integration into CRM, website, Telegram, n8nWired through our API or ready no-code scenarios; one key and a shared balance for every model.
- Payment by invoiceCompanies get an invoice for bank transfer, accounting documents and a top-up of the API balance.
- Support and limitsHelp with prompts and architecture, request limits raised individually for your load.
Inputs, limits and refunds
- Two shapes: Anthropic Messages (/v1/messages, x-api-key) and OpenAI chat/completions (Bearer)
- Streaming: Anthropic SSE with a billing event at the end; OpenAI chunks and data: [DONE]
- Input and output tokens are billed at their own per-1M prices (output costs more), minimum per call in min_rub
- A reserve by max_tokens (default 8192) is taken before the answer and settled after
- Failure before the first token — the reserve is refunded
- Image input only via the separate vision model (model: vision); images sent to the models on this page return 400 model_no_vision with nothing charged
- 30 requests per minute per key (all endpoints combined); model: smart picks the model tier itself
Error responses and refunds
# 401 {"type":"error","error":{"type":"authentication_error","message":"Invalid or revoked API key."}}
# 402 {"type":"error","error":{"type":"billing_error","message":"Not enough API balance: …"}}
# 429 {"type":"error","error":{"type":"rate_limit_error","message":"…"}} ← 30 requests/min per key
# 400 {"type":"error","error":{"type":"invalid_request_error","message":"Unknown model \"…\". See GET /v1/models."}}
# /chat/completions uses the OpenAI envelope: {"error":{"message":"…","type":"…","code":"insufficient_balance"}}
# streaming: the reserve (by max_tokens) is settled after the answer; a failure before the first token is refunded
# event: billing {"charged_rub":…,"reserved_rub":…,"balance_rub":…,"unit":"1M_tokens","rub_per_mtok_in":…,"rub_per_mtok_out":…,"rub_per_mtok":…}
Observed stability of the API and text models
Numbers come from the public /status summary: the share of successful requests over a period for the whole API and for the category as a whole (the summary has no per-model breakdown), no SLA promises. Failed generations are not charged, so retrying is safe.
Developer APIoperational
—now (2 h window)
74.5%last 30 days
74%last 90 days
Text & chatoperational
—now (2 h window)
99%last 30 days
94.5%last 90 days
Refunds and charges
- Failure before the first token — the reserve is refunded in full
- The max_tokens reserve is settled after the answer — only the real volume is charged
- In streaming the real charge arrives as the billing event
Tracking since Aug 8, 2026.
Use cases
1
Bulk product descriptions and SEO copy on the light model
2
Ticket classification and data labelling in a stream
3
Assistants over internal documentation and a company knowledge base
4
Agents with function calling for CRMs, spreadsheets and internal services
5
Translating and localising interfaces and content
6
Code generation and refactoring on the flagship
Prompting tips
- Split traffic by model: Flash for classification, extraction and templated copy, Max for reasoning, code and long context.
- Ask for JSON with an example schema — the result is easier to pass down the pipeline.
- For translation set the target language, tone and a glossary in the system prompt.
- Cap max_tokens: a short answer is cheaper and faster, and the reserve is smaller.
- Describe functions in tools concretely — name, purpose and typed parameters.
- For chat UIs enable stream: true so users see the answer right away.
Ready-made prompts for Qwen
Copy the prompt and the parameters into the request body — the fields are named exactly as in the /api/v1 contract. Replace the angle-bracket placeholders with your material.
SEO: title and description
For the page below suggest three title options under 60 characters and a description under 155 characters with the key phrase “<phrase>”: <page text>
max_tokens: 400Ticket classification
<the customer's message to classify>
system: Return JSON only: {"category": billing | delivery | quality | other, "priority": low | medium | high}max_tokens: 60Translation with localisation
Translate into German for an Austrian audience, keep formatting and placeholders in curly braces, adapt idioms: <text>
max_tokens: 1500Chatbot with a system prompt
<conversation history in messages, last turn from the user>
system: You are a consultant at a bicycle shop. Keep it short, ask about height and budget, suggest at most two models.stream: truemax_tokens: 500
Who the Qwen API is for
Industries where this kind of generation becomes part of the workflow rather than an experiment.
SaaS & appsAn AI assistant inside your product: OpenAI or Anthropic shape, Qwen via a swapped base URL.
Support & CRMTicket replies, classification, conversation summaries for managers.
E-commerceProduct descriptions, SEO copy, review replies — in bulk across the catalogue.
EdTechGrading, explanations at the learner's level, quiz generation.
Documents & legalDigests, version comparison, field extraction into structured data.
Media & contentArticle drafts, rewrites, headlines and translations — Qwen in the editorial pipeline.
Other models via the API
Same key, same balance, same request contracts.
FAQ
What is the Qwen API price per million tokens?
Each Qwen model has two prices per million tokens — input and output, output costing more — shown in the table above and in GET /v1/models, with the per-call minimum in min_rub. The light Flash is much cheaper than the Max flagship, so high-volume requests are best sent to it.
How do I connect the Qwen API in Node.js or Python?
Use the official OpenAI SDK: base_url zerocoder.com/api/v1, a zc-sk-… key and the Qwen model id from the table. The /v1/chat/completions format, the system, user and assistant roles and streaming match OpenAI. Python, Node.js and curl samples are in the “First request” block.
How do Qwen Max and Flash differ, and which should I choose?
Qwen Max is the flagship for reasoning, code and long documents; Qwen Flash is a fast light model for classification, extraction and templated copy at a lower price. A handy setup is Flash for the stream and Max for hard cases; both use one key with their own id from the table.
Which limits apply to the Qwen API?
A shared key limit of 30 requests per minute across all API endpoints; above it you get 429 rate_limit_error. max_tokens caps the answer and sets the reserve. The Qwen models on this page do not take image input — such a request returns 400 with nothing charged.
Is money refunded if the Qwen API returns an error?
Yes. If the Qwen answer never starts, the reserve is refunded in full automatically; format errors and rate limiting cost nothing. If a stream breaks midway only the generated tokens are charged, and the actual charge shows in the billing block of the response.
Can the Qwen API be paid by company invoice?
Yes. A legal entity requests an invoice in the API dashboard and pays by bank transfer; once credited, the amount is spent on Qwen and other models with one key. A card top-up is credited at once, and accounting documents are available on request. No Alibaba Cloud account needed.
Is there a trial period for the Qwen API?
There is no trial and no free API credit: the wallet opens with a minimum first top-up, then Qwen is charged per token. The cheapest way to judge quality is to run your own prompts on the light Flash model with a small max_tokens.
Do function calling and streaming work in the Qwen API?
Yes. On /v1/chat/completions Qwen accepts OpenAI tools and returns tool_calls, and with stream: true the answer arrives in chunks ending with data: [DONE], preceded by a billing block. The Anthropic format (/v1/messages) works too, but without tools.
