Qwen2.5 VL 72B Instruct API: toman pricing and code
qwen/qwen2.5-vl-72b-instruct
visionjsonYou are billed for the usage the request actually reported. Prices follow the market. Cached input: 116,050 toman / 1M.Pricing and top-ups
Qwen2.5 VL 72B Instruct example: reading an image into JSON
The example is picked from this model's own capabilities. Drop in your key and it runs as is.
# tip: a data URI works too — base64 the file and prefix it with data:image/jpeg;base64,
curl https://api.uttapen.ir/v1/chat/completions \
-H "Authorization: Bearer sk-up-…" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen/qwen2.5-vl-72b-instruct",
"messages": [{
"role": "user",
"content": [
{"type": "text", "text": "Return only the invoice number and the total, as JSON."},
{"type": "image_url", "image_url": {"url": "https://example.com/factor.jpg"}}
]
}]
}'import base64, json
from openai import OpenAI
client = OpenAI(base_url="https://api.uttapen.ir/v1", api_key="sk-up-…")
img = base64.b64encode(open("factor.jpg", "rb").read()).decode()
resp = client.chat.completions.create(
model="qwen/qwen2.5-vl-72b-instruct",
messages=[{
"role": "user",
"content": [
{"type": "text", "text": "Return the invoice number, the date and the total as JSON."},
{"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{img}"}},
],
}],
response_format={"type": "json_object"},
)
print(json.loads(resp.choices[0].message.content))import OpenAI from "openai";
import { readFileSync } from "node:fs";
const client = new OpenAI({ baseURL: "https://api.uttapen.ir/v1", apiKey: "sk-up-…" });
const img = readFileSync("factor.jpg").toString("base64");
const resp = await client.chat.completions.create({
model: "qwen/qwen2.5-vl-72b-instruct",
messages: [{
role: "user",
content: [
{ type: "text", text: "Return the invoice number, the date and the total as JSON." },
{ type: "image_url", image_url: { url: `data:image/jpeg;base64,${img}` } },
],
}],
response_format: { type: "json_object" },
});
console.log(JSON.parse(resp.choices[0].message.content));What is Qwen2.5 VL 72B Instruct good for?
Qwen2.5 VL 72B Instruct is one of Qwen (Alibaba)'s models. Put "qwen/qwen2.5-vl-72b-instruct" in the model field and the rest of your code stays as it is. Qwen2.5 VL 72B Instruct keeps 128,000 tokens in view at once, and that number is what caps the conversation history in your product. A single response can run to 115,200 tokens. On price it sits in the "cheap" band — cheaper than 135 and dearer than 271 of the other paid models in the catalogue.
Capabilities available on this id: it reads images directly, which makes it a real option for invoices, forms and screenshots; it returns schema-valid JSON through response_format, ready to hand to your code. All of it works through the standard parameters of the official OpenAI SDK — no custom client, no wrapper.
For a back-of-envelope figure: about 679 toman per 1,000-word exchange (232,100 in, 290,125 out, per 1M tokens). It supports cached input: repeated context is billed at 116,050 toman per 1M, which matters a lot if your system prompt is long. You are always charged for the usage the request actually reported, never for the estimate, and a failed request costs nothing.
The closest alternative with the same capabilities from a different provider is Sonar: Qwen2.5 VL 72B Instruct works out roughly 1.1× cheaper, and its context window is larger. Both run on the same key and the same code, so trying the other one is a single string change.
To make the figure concrete: 100,000 toman of credit buys roughly 147 thousand-word requests on Qwen2.5 VL 72B Instruct, and every 1,000 toman is about 1,473 words of round trip. A job with one million input tokens and one million output tokens comes to 522,225 toman in total. Filling this model's 128,000-token window costs 29,709 toman on the input side alone, which is the real reason to keep conversation history short.
Qwen (Alibaba) has 53 models in our catalogue; the cheapest is Qwen3.7 Flash at 60 toman per thousand words and the dearest Qwen3.8 Max (0902) at 3,017. Among the less common parameters it accepts logit_bias, logprobs, repetition_penalty, top_k, top_logprobs — all through the standard request body. It does not support include_reasoning, reasoning, tool_choice, tools, which most models here do, so test before switching if your code relies on them. By context size the nearest option from another provider is Nova Micro 1.0 at 128,000 tokens.
What to use it for: high-volume work such as classification, tagging and bulk summarising, pulling text and fields out of images, extracting data against a fixed schema.
Qwen2.5-VL is proficient in recognizing common objects such as flowers, birds, fish, and insects. It is also highly capable of analyzing texts, charts, icons, graphics, and layouts within images.
Three real jobs, priced on this model
Each figure is derived from the prices above and moves when they do.
| Job | Tokens | Cost |
|---|---|---|
| One chat turn with a medium history | 1,500 in + 400 out | 464 toman |
| Summarising a ten-page document | 4,000 in + 600 out | 1,102 toman |
| Classifying a thousand short rows | 120,000 in + 20,000 out | 33,655 toman |
Frequently asked
- How do I call Qwen2.5 VL 72B Instruct from Iran?
- Sign up with your mobile number, top the wallet up in toman, create an API key, then in the official OpenAI SDK point base_url at https://api.uttapen.ir/v1 and set model to "qwen/qwen2.5-vl-72b-instruct". Nothing else in your code changes.
- What does Qwen2.5 VL 72B Instruct cost in toman?
- 232,100 toman per 1M input tokens and 290,125 toman per 1M output tokens; a 1,000-word request is around 679 toman. You pay for the usage that request actually reported, and a failed request costs nothing.
- How much input does Qwen2.5 VL 72B Instruct take?
- Up to 128,000 tokens per request, roughly 96k words. A single answer can reach 115,200 tokens.
- Does Qwen2.5 VL 72B Instruct support streaming and image input?
- Streaming (stream=true) works on every model here. Image input is accepted through image_url, as a data URI or a public URL. Structured output through response_format works too.