Qwen-Audio 3.1 price cuts: rebuild your voice bill with three numbers

Alibaba's Qwen-Audio 3.1 price cuts are reported as about 70% for text-to-speech, 85% for Realtime and up to 95% for speech recognition. TBreak gives percentages only, with no absolute prices, so you cannot rebuild a bill from the article alone. You need three numbers of your own: monthly volume, the price you pay today, and the cut that applies to that product.

The new bill is the old bill multiplied by one minus the cut. That is simple arithmetic, and it is also the part people skip when a headline says 95%. Note the words up to: the largest cut applies to one product tier, not to every call.

Apply the percentages to your own bill

The rows below are an illustration of the arithmetic on a round $100 monthly spend. They are not Qwen prices. Replace the $100 with your own figure.

• Use the invoice from the last full month, not a list price.
• Check whether the cut is a list-price change or a promotion with an end date.
• Re-run the numbers when the vendor publishes absolute prices.

Compare with a flat, per-character price

A cheaper rate only helps if the quality fits and the access works for you. Sume's TTS Router prices Cartesia Sonic by character: Cartesia's list price of $38 per million credits (about one credit per character) times Sume's 1.25 margin, which is about $0.0475 per 1,000 characters, rounded up to the cent for each job. Sume STT is about $0.01 a minute. A 700-character voiceover therefore costs about 3.3 cents, which rounds to 4 cents on a job. Read the Sume TTS Router catalog for the live figures.

Rebuild your bill in four steps

• Export last month's usage by product: characters for TTS, audio minutes for ASR, audio tokens or minutes for Realtime.
• Multiply by the old unit price to confirm it matches the invoice.
• Apply the reported cut and flag any product the vendor has not priced in absolute terms.
• Add the same volume at Sume's per-character and per-minute prices to see the gap.

import os, requests

r = requests.post(
    "https://api.sume.com/v1/tts-router/generate",
    headers={
        "Authorization": f"Bearer {os.environ['SUME_API_KEY']}",
        "Idempotency-Key": "tts-demo-001",
    },
    json={
        "model": "sonic-3.6",
        "transcript": "Welcome back. Today we compare three prices.",
        "voice": {"id": os.environ["SUME_VOICE_ID"]},
        "timestamps": {"words": True},
    },
    timeout=30,
)
r.raise_for_status()
print(r.json())

What Sume does not do

Sume does not route to Qwen models. The TTS Router lists Cartesia Sonic only, and STT is one hosted endpoint with a 10-minute limit and no streaming. I could not read the marktechpost article about Qwen-Audio 3.1 Realtime because the fetch was refused, so nothing here depends on it. Treat the percentages as reported by TBreak until Alibaba publishes the price sheet.

Related reading

For a wider comparison of what voice costs at scale, see the up-to-213x voice API cost gap and the price per 1,000 characters ranking.