Qwen 3.8 Flash via API: setup, pricing and bulk text tasks

Qwen 3.8 Flash is the junior and cheapest model of the Qwen 3.8 line in the Zukko catalog. It fits text tasks at volume: summarisation, field extraction, labelling. Below is how to connect the qwen 3.8 flash api, what it costs at real volume, where the model's limits are, and how to tell whether it suits your workload.

Qwen 3.8 Flash pricing and specs

Prices in USD per 1M tokens, as of 2026-10-06.

For Qwen a cache write is not charged separately, and a cache read is counted if the provider reports it in the response.

The vendor's official price for Qwen 3.8 Flash is $0.15 / $0.47. The intelligence index is the Artificial Analysis Intelligence Index, a composite benchmark score: 40 for Flash, 45 for Max.

What this means in practice:

• Text only. Qwen 3.8 Flash does not accept image input. If you need images, the line has Qwen 3.8 Max.
• Caching doesn't lower the price. A cache read costs the same as regular input, $0.038, and a write is not charged separately. A repeated system prompt is paid in full on every request — but there is no write premium either.
• Long context is cheap. A 500,000-token document as input costs 500,000 × 0.038 / 1,000,000 = $0.019.

For the current model list and prices, call GET https://api.zukko.pro/v1/models or open the Models page in the dashboard.

How to connect Qwen 3.8 Flash via API

Create a fetch_... key in the dashboard; the base URL for the OpenAI-compatible format is https://api.zukko.pro/v1. If this is your first connection, start with the quick start↗.

curl

curl https://api.zukko.pro/v1/chat/completions \
  -H "Authorization: Bearer fetch_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.8-flash",
    "messages": [
      {"role": "system", "content": "Shorten the text to three sentences."},
      {"role": "user", "content": "Text to shorten..."}
    ]
  }'

Python: extracting fields as JSON

import json
from openai import OpenAI

client = OpenAI(base_url="https://api.zukko.pro/v1", api_key="fetch_your_key")

PROMPT = (
    "Extract the fields brand, color, size from the product description. "
    "Return only a JSON object, no explanations. Use null for missing fields."
)

resp = client.chat.completions.create(
    model="qwen3.8-flash",
    messages=[
        {"role": "system", "content": PROMPT},

The model's answer is text, so validate the JSON: at volume some responses may fail to parse, and those should be retried.

Python: streaming

For long answers it is handy to receive text in pieces. At the end of the stream the server sends a separate chunk with usage and an empty choices, so the check is required:

stream = client.chat.completions.create(
    model="qwen3.8-flash",
    messages=[{"role": "user", "content": "Summarise this report: ..."}],
    stream=True,
)
for chunk in stream:
    if chunk.choices:
        print(chunk.choices[0].delta.content or "", end="", flush=True)

More on request parameters in the chat/completions docs↗.

What bulk processing on Qwen 3.8 Flash costs

Example: summarising 10,000 documents. Assumptions: each request uses 3,000 input tokens (instruction plus document text) and 300 output tokens (the summary).

Formula:

For Qwen 3.8 Flash:

• input: 10,000 × 3,000 = 30,000,000 tokens → 30 × $0.038 = $1.14
• output: 10,000 × 300 = 3,000,000 tokens → 3 × $0.088 = $0.264
• total: $1.404

The same volume on neighbouring models:

The honest reading of the table: on price and index, Qwen 3.8 Flash sits in the middle of the low-cost group, not at the edge. Gemini 3.8 Flash is cheaper at this volume with a similar index, and DeepSeek V4.1 Flash is cheaper still. That is a reason to compare models on your own data rather than pick by name. GLM 5.3 Flash has a 128K context, so it won't take very long documents, while Qwen 3.8 Flash has 1M.

What to use Qwen 3.8 Flash for

Reasonable scenarios:

• Summarising and shortening texts, where the answer is much shorter than the input.
• Field extraction from descriptions, emails, requests — with JSON validation in code.
• Labelling and classification with short answers.
• Long documents that don't fit into 128K.
• Pairing with Qwen 3.8 Max: the main stream on Flash, and the requests Flash failed in your tests on Max, in one format with one key.

What not to expect: image handling (use Max or another model with image input) and savings from caching.

How to check the model handles your task

The intelligence index is a general guide; quality on your task can only be checked with your examples:

1. Collect 50–100 real inputs with reference answers.
2. Run them through Qwen 3.8 Flash and one or two neighbouring models from the table.
3. Count the share of correct answers and the actual cost from usage.
4. Pick the cheapest model whose accuracy satisfies you.

For 100 examples at 3,000 input and 300 output tokens, a run on Qwen 3.8 Flash costs 100 × (3,000 × 0.038 + 300 × 0.088) / 1,000,000 = 100 × 0.0001404 = about $0.014.

For a comparison with GLM, see Qwen vs GLM↗; for DeepSeek, Qwen and GLM side by side, see the article on Chinese models via API↗. To get started, create a key in the dashboard and do a test run.

FAQ

Does Qwen 3.8 Flash accept images?

No, text only. For images, the Qwen line has qwen3.8-max, which supports image input.

Is caching worth it for Qwen 3.8 Flash?

Price-wise, no: a cache read costs $0.038, exactly the same as regular input. A long system prompt is paid in full on every request, so account for it in your estimate.

How does Qwen 3.8 Flash differ from Qwen 3.8 Max?

Per the catalog: Flash is cheaper ($0.038 / $0.088 vs $0.1 / $0.263), has a lower intelligence index (40 vs 45), and no image input. Both have a 1M context.

Can I call Qwen 3.8 Flash through the Anthropic SDK?

Yes, the gateway accepts any model in both formats. For bulk scripts, the OpenAI SDK with base URL https://api.zukko.pro/v1 is usually simpler.