Python TTS cost calculator: Sume job rounding vs per-character rates

Per-million prices hide a billing detail. Sume rounds each job up to the cent, with a one-cent minimum, so many tiny jobs cost more than one job of the same total length. Microsoft lists MAI-Voice-2.1 at $22 and Flash at $15 per million characters (Digital Applied, read 2026-10-06); the function below applies those as flat rates and does not model any rounding Microsoft might apply.

The function takes a list of job lengths in characters, so it answers both "one big job" and "one job per line".

The calculator

def sume_cents(chars):
    return max(1, -(-chars * 19 // 4000))   # $0.0475 per 1,000, ceil to cent

def flat(chars, per_million):
    return chars * per_million / 1_000_000

def report(jobs):
    total = sum(jobs)
    print("chars", total, "jobs", len(jobs))
    print("sume  $%.2f" % (sum(sume_cents(j) for j in jobs) / 100))
    print("mai   $%.4f" % flat(total, 22))
    print("flash $%.4f" % flat(total, 15))

report([900])
report([90] * 10)
report([40] * 100)

What it shows

Takeaway

Batch lines into fewer jobs when audio is going to be one narration anyway. A hundred 40-character jobs cost $1.00 on Sume because each is rounded to the one-cent minimum, while the same 4,000 characters in one job is $0.19. The only constraint is the 20,000-character cap per job.