AI API pricing units decoded: credits, tokens, characters, hours

Vendors do not agree on what to count. Runway counts credits, Google counts tokens, ElevenLabs counts characters, Microsoft counts characters and hours, and fal counts seconds. To compare anything you convert everything to dollars per the unit you actually consume: per second of video, per minute of speech, per image.

Sume's side is simple. Sume reserves against a spendable USD balance (see the billing and credits docs), so there is no credit conversion step on a Sume bill.

Unit by unit

A converter

Three lines of arithmetic cover most pages. Run this with Python 3 to check your own figures.

# Convert vendor price units to dollars (rates read 2026-10-05).
CREDIT_USD = 0.01          # Runway: 1 credit = $0.01

def runway_per_second(credits_per_second):
    return credits_per_second * CREDIT_USD

def per_million_chars_to_1k(usd_per_million):
    return usd_per_million / 1000

def per_hour_to_per_minute(usd_per_hour):
    return usd_per_hour / 60

print("Runway Gen-4 Turbo $/s:", runway_per_second(5))
print("MAI-Voice-2.1 $/1k chars:", per_million_chars_to_1k(22))
print("Scribe v2 $/min:", round(per_hour_to_per_minute(0.22), 5))

Habits that prevent mistakes

Convert before you compare, and write the unit in every row of your spreadsheet. A price per million characters looks enormous next to a price per second until you convert both to a finished minute.

Note the date. Several of the figures above are promotional or time limited: ElevenLabs v4 promotional pricing runs until October 12, and the Microsoft streaming rate is for the rest of the year. Stamp every figure with the day you read it, as these tables do.

On Sume, for per-second models the price is the provider list price times 1.25; the Sume API rate card lists the fixed prices.

• Credits x $0.01 for Runway.
• Per million divided by 1,000 gives per 1,000 characters.
• Per hour divided by 60 gives per minute.