What does "per 1M tokens" mean
Once you understand this unit, you can price any model for any use case in under a minute
What does "per 1M tokens" mean
It is a pricing unit. Like kilowatt-hours for electricity, or per-kilometre for fuel. A way to compare costs at a standard scale.
$3.00 per 1M tokens means: process one million tokens, pay $3.00. Process 500,000 tokens, pay $1.50. Process 10 million tokens, pay $30.00. The rate is constant. The cost scales with usage.
The reason the industry settled on one million as the unit is that individual API calls are tiny — fractions of a cent. Pricing at $0.000003 per token is technically accurate but practically unreadable. Per million keeps the numbers human.
The number that makes it real
One typical API call — a 200-word prompt, a 150-word response — uses roughly 460 tokens. At $3.00/1M, that call costs $0.00138. Under a cent and a half.
At 100,000 calls per month: $138. At 1,000,000 calls per month: $1,380.
The per-million rate is the same. Your cost scales with your volume.
Why this matters to you
This unit is the key to evaluating any model for any use case. Once you know three numbers — your average call size in tokens, your monthly call volume, and the model's price per million — you can calculate your monthly cost in thirty seconds.
Most developers skip this until their first invoice. The ones who do it upfront make better model choices and have no surprises.
The thirty-second calculation
1. Average prompt length in words ÷ 0.75 = input tokens per call 2. Average response length in words ÷ 0.75 = output tokens per call 3. (Input tokens × input price/1M) + (output tokens × output price/1M) = cost per call 4. Cost per call × monthly call volume = monthly cost
Run this against two or three models before you pick one. The result sometimes surprises people.
Verified March 2026 · Source: Anthropic, OpenAI, Google pricing documentation