Qwen 3.8 Omni Flash

Qwen 3.8 Omni Flash is the Qwen built to take in more than text. You send speech, images or video with your prompt and it answers in text: a transcript, a summary of a call, the steps shown in a screen recording. It shares its design with Qwen 3.8 Flash and holds up to a million tokens in one request.

Sound is inexpensive to send. Alibaba counts seven tokens for each second of audio, so an hour-long recording is about 25,000 tokens of input. It thinks before answering by default, with adjustable effort.

• Transcribing meetings, calls and voice notes.
• Summarising long recordings and answering questions about them.
• Reading screen recordings and short videos.
• Pulling names, dates and actions out of speech.

• Replies come back as text. For a spoken answer, pair it with a text-to-speech model on the same key.
• Thinking is on by default. Turn the effort down for straightforward transcription to keep replies quick.

For scale: 1,000 requests of about 3,000 tokens in and 1,000 out cost $1.20 — a long prompt and a page of answer each. Every rate is published and stays the same from one request to the next. See every model.

Use the OpenAI SDK you already have — point it at Roar AI and name the model. Read the docs.

from openai import OpenAI

client = OpenAI(base_url="https://api.roar-ai.com/v1", api_key="roar_live_…")

reply = client.chat.completions.create(
    model="qwen3.8-omni-flash",
    messages=[{"role": "user", "content": "Summarise this contract in plain English."}],
)