AI TOKEN + INFERENCE COST CALCULATOR

Token Burn

A meter for the invisible fuel behind AI work. Enter what you know, estimate what you do not, and keep the assumptions attached to the number.

All currency math resolves through integer micro-dollars. The catalog snapshot and calculation state are preserved in every receipt.

Calculated API-equivalent cost

$0.00

Loading verified catalog…

Fresh + cached + cache write + output + tools

Model

Service

Context tier

01 / KNOWN TOKEN MODE

Set the burn

Total tokens are only the start. The input/output mix and eligible cache use determine the meter.

Total tokens 1K1M1B10B

Cached input % Fresh input % Cache writes % Output + reasoning %

Demonstration assumption: 70% cached input, 20% fresh input, and 10% output. This is not observed usage.

Advanced token fields exact counts override the composition mix

Fresh input Cached input Cache writes Output + reasoning Fixed tool charges (USD)

Daily—

Weekly—

Monthly—

Annual—

02 / MODEL COMPARISON

Same workload. Different meter.

Sort comparison

Price per million is not cost per successful task. Capability, retries, token efficiency, tool reliability, latency, and human review change outcome cost. Open-weight availability is not a universal hosting price.

03 / WORKLOAD COMPOSER

Estimate recognizable work

Ranges expose uncertainty. They do not pretend an estimate is measured usage.

Workload preset

Documents Pages or slides each Words per page Populated cells Source lines of code Agent turns Analysis passes Revision cycles Expected response tokens Context replay % Cache reuse % Subagents Retry / repair cycles Images, charts, tables / doc Scanned-document classification

Estimated token range

Low—

Expected—

High—

Assumptions used

Expected estimate across models

04 / IMAGE COST LAB

Price the iterations, not just the keeper

Generating exactly one image and accepting it unchanged is rarely a credible planning assumption.

Image model

Accepted final images Variants per accepted image

Resolution

Quality

Service

Operation

Reference images Prompt tokens / variant Image input tokens / variant

Estimated image meter

—

Generated—

Per generated image—

Per accepted image—

Total input overhead—

Text input overhead—

Reference-image input—

Representative official output prices across image models

Accepted-image and variant counts remain fixed. Resolution and quality labels stay visible because model outputs are not assumed equivalent.

05 / SUBSCRIPTION LENS

API-equivalent value

This compares meters. It does not claim an unlimited token entitlement, financial benefit, or realized return.

Subscription price

Billing period

Active usage days Usage tokens Codex credits used Overage purchases

API-equivalent / subscription meter —

API-equivalent value—

Effective subscription / 1M—

Break-even usage—

Monthly subscription—

Annualized subscription—

Official Codex credits—

API list price is not provider marginal cost, a promised plan entitlement, or proof that this usage would have been purchased through an API. Credits remain separate from dollars.

06 / INTERPRET THE NUMBER

The meter is one part of the economics

Composition matters

A billion cached input tokens, fresh input tokens, and output tokens produce radically different costs.

Success changes the denominator

A cheaper model that retries, drops tools, or needs more human repair can cost more per useful outcome.

Deployment types are distinct

Direct API, third-party hosting, subscription plans, local inference, and open-weight availability are different economic arrangements.

Estimates need provenance

Workload estimates are planning ranges. Historical receipts preserve the catalog and assumptions that produced them.

07 / PRICING PROVENANCE

Every rate carries its source

08 / RECEIPTS + OFFLINE

Keep the assumptions attached

Exports preserve the selected catalog snapshot, model, tier, composition, formula, result, and interpretive limits.

Download offline HTML

Receipt versiontoken-burn-receipt.v1 Catalog— Verified—