Qwen API: every Qwen model on one key
Every Qwen model Qwen ships to its API, 10 on Revana today, through one OpenAI-compatible endpoint. Point any OpenAI SDK at https://api.revana.si/v1, name the model, and pay per token from prepaid credit. A request that gets no answer costs $0.00.
Qwen3.8-Max $1.92 in / $5.76 out per 1M tokens
Qwen3.8-Flash $0.1646 in / $0.5555 out per 1M tokens
Qwen3.7-Flash $0.0412 in / $0.1646 out per 1M tokens
Revana prices as of November 2026.
Every Qwen model and what it costs
USD per million tokens on Revana as of November 2026. Click a model for its full price sheet, cache-write rates and code. Reasoning and thinking tokens bill as output.
Qwen3.8
The newest Qwen models, each with a context window of about 1M tokens. 4 models on Revana, USD per million tokens.
| Model | Input | Cached input | Output | Context |
|---|---|---|---|---|
Qwen3.8-Maxqwen3.8-max | $1.92 | $0.24 | $5.76 | 1M |
Qwen3.8-Flashqwen3.8-flash | $0.1646 | $0.0206 | $0.5555 | 984K |
Qwen3.8 27Bqwen3.8-27b | $0.6172 | $0.1235 | $2.4686 | 984K |
Qwen3.8 2.4T-A95Bqwen3.8-2.4t-a95b | $2.4686 | $0.3086 | $7.4058 | 984K |
Qwen3.7
The previous release, with the lowest-cost Flash model on the list. 3 models on Revana, USD per million tokens.
| Model | Input | Cached input | Output | Context |
|---|---|---|---|---|
Qwen3.7-Plusqwen3.7-plus | $0.4116 | $0.4116 | $1.6463 | – |
Qwen3.7-Maxqwen3.7-max | $2.4685 | $2.4685 | $7.4054 | – |
Qwen3.7-Flashqwen3.7-flash | $0.0412 | $0.0083 | $0.1646 | 984K |
Qwen3.6
Kept for prompts already tuned to it. 2 models on Revana, USD per million tokens.
| Model | Input | Cached input | Output | Context |
|---|---|---|---|---|
Qwen3.6-Plusqwen3.6-plus | $0.4116 | $0.4116 | $2.4694 | – |
Qwen3.6-Flashqwen3.6-flash | $0.2469 | $0.2469 | $1.4809 | – |
Other Qwen models
1 model on Revana, USD per million tokens.
| Model | Input | Cached input | Output | Context |
|---|---|---|---|---|
Qwen4-Maxqwen4-max | $2.5715 | $0.1286 | $7.7143 | 992K |
Match the Qwen model to the job
What a workload costs on Revana at today’s prices, worked out from the token rates above.
Very high volume
100,000 calls that each read 1,000 tokens and write 200: Qwen3.7-Flash: $7.41 · Qwen3.8-Flash: $27.57. Qwen3.7-Flash costs 73% less for this job. Tagging, routing and short rewrites where cost per call matters most.
A middle option
1,000 requests that read 4,000 tokens and write 800: Qwen3.8 27B: $4.44 · Qwen3.7-Plus: $2.96. Qwen3.7-Plus costs 33% less for this job. Priced between Flash and Max, for requests Flash gets wrong.
Long documents
1 question about a 300K-token document: Qwen3.8-Flash: $0.0502 · Qwen3.8-Max: $0.58. Qwen3.8-Flash costs 91% less for this job. Repeat questions read the document from the cache at the cached rate.
Hard agent steps
100 agent steps that read 20K tokens (10K cached) and write 3K: Qwen3.8-Max: $3.89 · Qwen3.8 2.4T-A95B: $5.00. Qwen3.8-Max costs 22% less for this job. Send the hard steps to Max and keep routine ones on Flash.
Call Qwen with the code you already have
Two settings change: the base URL and the key. Messages, tools, response_format and streaming stay in the OpenAI format, and switching to another Qwen model, or another maker’s, is a change to the model field.
curl https://api.revana.si/v1/chat/completions \
-H "Authorization: Bearer $REVANA_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"qwen3.8-flash","messages":[{"role":"user","content":"Write a haiku about routing."}]}'import os
from openai import OpenAI
client = OpenAI(base_url="https://api.revana.si/v1", api_key=os.environ["REVANA_API_KEY"])
reply = client.chat.completions.create(
model="qwen3.8-flash",
messages=[{"role": "user", "content": "Write a haiku about routing."}],
)
print(reply.choices[0].message.content)
print(reply.usage.cost_usd) # what this request cost, in USDimport OpenAI from "openai"
const client = new OpenAI({ baseURL: "https://api.revana.si/v1", apiKey: process.env.REVANA_API_KEY })
const reply = await client.chat.completions.create({
model: "qwen3.8-flash",
messages: [{ role: "user", content: "Write a haiku about routing." }],
})
console.log(reply.choices[0].message.content, reply.usage.cost_usd)The answer is OpenAI’s chat.completion object, and usage.cost_usd is the exact amount taken from your balance. For 3,000 tokens in and 500 out on Qwen3.8-Flash:
{
"id": "chat_3f9c…",
"object": "chat.completion",
"model": "qwen3.8-flash",
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": "…"
},
"finish_reason": "stop"
}
],
"usage": {
"prompt_tokens": 3000,
"completion_tokens": 500,
"total_tokens": 3500,
"prompt_tokens_details": {
"cached_tokens": 0
},
"cost_usd": 0.000772
}
}Send "stream": true for server-sent events, and "stream_options": {"include_usage": true} to get the cost in the last chunk. Every field is in the chat completions docs.
Pay only for the answers you get
- Automatic fallback. A model with more than one route starts on its lowest-cost route; a busy or failed route hands the request to the next before any answer is sent. Each model page shows its routes.
- Held, then charged. While a request runs, Revana holds its prompt plus
max_tokens(8,192 output tokens when you set none). When it ends the hold is released and only the tokens used are charged. - No answer costs $0.00. If every route fails you get an OpenAI-shaped error and no charge;
503 model_busyis never billed, so a short retry is safe. - Backup routes stay under the hold. When a model’s listed route is busy, a backup can cost more; you are charged that route’s cost, never more than the hold, and it is always in
cost_usd.
Make images and video from Claude Code, Cursor and ChatGPT
- Connect the Revana MCP server to Claude Code, Claude Desktop, claude.ai, Cursor, VS Code, Codex CLI or ChatGPT, and your agent can quote, generate and download images and video on your credit.
- Every language model and its price is published as llms.json and through
GET /v1/models, so code can pick a model by price at run time. - Any tool that sends HTTP works too, such as n8n’s HTTP Request node:
POST https://api.revana.si/v1/chat/completionswith your key as a Bearer token.
Qwen Image and Wan video on the same balance
Alibaba's Qwen Image and Wan video models run on the same Revana key through POST /v1/generations, from the same credit as your Qwen tokens.
| Model | Kind | From |
|---|---|---|
| Z Image | image | $0.01 an image |
| Z Image Turbo | image | $0.02 an image |
| Qwen Image | image | $0.03 an image |
| Qwen Image 2.0 | image | $0.03 an image |
| Wan 2.6 Flash | video | $0.03 a second |
| Wan 2.6 I2V Flash | video | $0.03 a second |
| Wan 3.0 | video | $0.04 a second |
| Wan 2.2 Turbo | video | $0.05 a second |
Every image, video and audio model and its price is at revana.si/models. A generation that fails costs nothing.
From signup to your first Qwen call
- Create a key. Sign up and create an API key in the dashboard. It works with every language, image and video model.
- Add prepaid credit. Packs start at $10 and 100 credits are $1. Credit never expires; no subscription or monthly minimum.
- Send the sample request. Copy the curl above, set your key, and read the answer and its
cost_usd.
FAQ
How do I get a Qwen API key?
Sign up at revana.si, create a key in the dashboard and add prepaid credit. The same key calls every Qwen model with any OpenAI SDK at https://api.revana.si/v1; no Alibaba Cloud account is needed.
Which Qwen model costs the least?
Qwen3.7-Flash at $0.0412 in and $0.1646 out per million tokens on Revana (prices as of November 2026).
Can Qwen generate images through Revana?
Yes, with the Qwen Image models on POST /v1/generations, priced per image and paid from the same balance. The chat models themselves answer in text.
How is Revana different from Alibaba Cloud Model Studio?
Revana is one OpenAI-compatible endpoint and one prepaid balance for Qwen, Claude, GPT, Gemini and the image and video models, with cost_usd on every answer and automatic fallback where a model has more than one route.
Is it the same model I would get from Alibaba?
Yes. Revana passes your request to the Qwen model unchanged and returns its answer in the OpenAI format, with the usage it reported.
Which features does the chat endpoint leave out?
Text, pictures as image_url and function tools are supported. Web search, audio, files, video input and built-in tools are refused with 400 unsupported_parameter.
Is there a subscription?
No. Pay per token from prepaid credit, packs start at $10, credit never expires, and a request that gets no answer costs nothing.
Start with one API key
One key for every language, image, video and audio model on Revana. Prepaid credit, no subscription, and a request that fails costs nothing.