Qwen3.8 Flash-Next

Free Qwen API

Zero token charges. No deposit. An OpenAI-compatible API for text generation, ready for your application.

Your model. One API key.

Input, output, and cached-input tokens. No prepaid balance is required for this model.

250 tok/s

Median decode. Actual throughput depends on the request and server load; it is not a latency guarantee.

Shared capacity is subject to availability. If the service is busy, retry later; a free Qwen request is not silently switched to a paid model.

Make your first request

Sign in to nafnaf and create an inference API key.
Set NAFNAF_API_KEY in your server's environment or secret manager. Keep it out of browser code, repositories, and prompts.
Send a Chat Completions request to https://inference.nafnaf.ai/v1 with model qwen3.8-flash-next.

Qwen curl request

curl --fail-with-body https://inference.nafnaf.ai/v1/chat/completions \
  -H "Authorization: Bearer $NAFNAF_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.8-flash-next",
    "messages": [{"role": "user", "content": "Explain GPU memory in one sentence."}],
    "max_tokens": 256,
    "stream": false
  }'

For Python, install the openai package and point its client at nafnaf:

Qwen Python request

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["NAFNAF_API_KEY"],
    base_url="https://inference.nafnaf.ai/v1",
)

response = client.chat.completions.create(
    model="qwen3.8-flash-next",
    messages=[{"role": "user", "content": "Explain GPU memory in one sentence."}],
    max_tokens=256,
)
print(response.choices[0].message.content)

To stream tokens, set stream to true. Qwen uses Chat Completions; do not send it to the Responses endpoint. The live catalog lists supported protocols for each model.

Before you build

Do I need to rent a GPU?

No. This is a hosted model API. Rent a GPU workspace only when you need your own model, runtime, or training job.

Does free Qwen make my whole API key free?

No. An inference key can access other models too. GLM and MiMo have their own published rates. Pin the model ID in your application and do not configure an automatic paid fallback unless you intend to use it.

What happens when a request fails?

Check the HTTP status. A 401 means authentication failed; 429 means a limit or shared capacity was reached. Back off before retrying and honor a Retry-After header when supplied. A 5xx response can mean the model is temporarily unavailable.

Live model catalog · Inference API specification · Privacy policy