Bitcent: The 10⁻¹⁰ Dollar Unit for AI Pricing

I’ve been playing with the xAI API recently, and noticed something interesting about their pricing model. The API returns usage costs as integers, but the numbers didn’t immediately make sense.

For example, a Grok Imagine generation costs 330,000,000. On their pricing page, that lists as $0.033. Or take grok-4-1-fast-reasoning, which returns prices like:

"promptTextTokenPrice": "2000",
"cachedPromptTokenPrice": "500",
"completionTextTokenPrice": "5000"

Represented as follows on the Models page:

Input: $0.20
Cached input: $0.05
Output: $0.50

Doing the math, that means each unit is 1/10,000,000,000 (one ten-billionth) of a dollar ($10^{-10}$). I needed a name for this conceptual unit to keep things sane, so I started calling it a bitcent.

I’ve no idea what other folks are calling this unit of measure, but there seemed to be little concensus after a quick Grok search.

The problem: Dust Pricing

We are in the era of “dust pricing”. API calls often cost a microscopic fraction of a penny. Dealing with this usually leads to one of two bad places:

1. Floating point dollars: $0.00000001 — Hello rounding errors and scientific notation confusion.
2. Naming confusion: “Microdollars”? “Nanodollars”? “Token-cents”?

xAI’s approach of using an integer with a fixed precision is actually the right way to handle currency (as any fintech dev will tell you). But we need a shared vocabulary for it.

Introducing the Bitcent

The bitcent represents exactly $10^{-10}$ dollars.

This is what xAI is already using

Why that specific number? It’s the sweet spot for avoiding floating point math while covering the vast range of AI pricing.

• 1 Bitcent = $0.0000000001
• 1 Cent = 100,000,000 Bitcents
• 1 Dollar = 10,000,000,000 Bitcents

This allows us to express prices that are currently “dust” as clean, readable integers:

Why “Bitcent”?

I considered other names, but they all fall short:

• Nanodollar: Technically $10^{-9}$ ($0.000000001). It maps poorly to the $10^{-10}$ scale xAI is using.
• Pico-something: Too small ($10^{-12}$).
• Tokencent (or token-cent): a bit long, and “price per token” will typically be in the same sentence, too many “token” words.
• Neurcent (or neurocent): interesting reference to “neural” nets, but perhaps too pretentious?

“Bitcent” works because it hints at digital currency (“bit”) while grounding it in familiar financial terms (“cent”). It sounds like a natural subdivision of modern digital money.

If we want to standardize how we talk about AI costs across providers, adopting the bitcent as the standard unit for $10^{-10}$ USD seems like a solid move. It turns “0.000025 cents” into “2,500 bitcents”—a number humans can actually read and compare.

For example, these are the current (as of Jan/2026) prices for some leading LLMs:

As prices continue to plummet, I think bitcents will make even more sense!