Estimate a Sume TTS bill in Python before you submit: count characters

The formula

The cost of a Sume TTS job is the number of characters in the transcript times $0.0475 divided by 1,000. A 5,000-character script is 5 x $0.0475 = $0.2375. Spaces and punctuation count, so measure the exact string you will send.

A request can carry 1 to 20,000 characters, but a job also fails with tts_duration_exceeded if the audio runs past 1,200 seconds. That second limit usually bites first.

Examples

Use the table to sanity-check your function.

The function

The code uses Decimal so the arithmetic does not pick up float noise. It warns when a script is likely to exceed 1,200 seconds, using Cartesia's published rule of thumb that a minute of audio is about 750 to 800 characters.

from decimal import Decimal

RATE = Decimal("0.0475")      # USD per 1,000 characters
MAX_CHARS = 20000
CHARS_PER_MIN_LOW = 750       # Cartesia pricing page rule of thumb

def estimate(text):
    n = len(text)
    if not 1 <= n <= MAX_CHARS:
        raise ValueError(f"{n} characters is outside 1-{MAX_CHARS}")
    cost = (Decimal(n) * RATE / 1000).quantize(Decimal("0.0001"))
    minutes = n / CHARS_PER_MIN_LOW
    return cost, minutes

if __name__ == "__main__":
    script = "Welcome to the Q4 review. " * 200
    cost, minutes = estimate(script)
    print(f"${cost} for about {minutes:.1f} minutes at the slow end")
    if minutes > 20:
        print("Split this script: a job over 1,200 seconds fails")

Using it well

Cartesia's pricing page says one minute of audio is about 750 to 800 credits, with a credit being one character. The rule is for planning only. Real timing depends on voice, speed and punctuation. If your estimate is near 15,000 characters, split the script into chapters or scenes and submit each separately. The same code then tells you the cost of every part, and you can store it next to the job id.

Python's len counts code points, which is what most scripts are made of. If you use emoji or combining marks heavily, test one sample against the job's billed usage before you rely on the estimate.

• Strip stage directions you do not want spoken before you count.
• Estimate per scene, not just per script, to find the expensive part.
• Log the estimate and compare it with the final invoice once a month.
• Prices change; read the current rate from GET /v1/catalog if you automate budgets.

Putting it in a pipeline

Call the estimate before every submit and refuse to send a script that would cost more than a limit you set, for example $2 for a single job. That keeps a pasted novel from becoming a surprise charge. Log the job id, the character count and the estimate in the same line, so a later invoice check is one search.

For a batch of slides or announcements, estimate each item and sum them. A deck of 20 slides at 600 characters is $0.57, and the code should print exactly that. If the printed value differs from your hand arithmetic, fix the code before it touches a paid route.

Finally, remember that the estimate is a ceiling for planning. Billing follows the characters the API receives, and a request that fails validation before it runs does not create audio. Retries with the same Idempotency-Key do not create a second charge, which is why you should always send one.