Your goal: inspect the arithmetic
I is total native input; R is cache reads; W5/W1 are five-minute/one-hour writes; O is billed output including reasoning. Counts are nonnegative integers; cache categories are disjoint subsets of input.
Uncached input U = I − R − W5 − W1
Supported basic/read cost C = U × input_rate + R × read_rate
+ O × output_rate + request_fee
Repeated profile cost = N × C
Combined workload = sum of complete profile costsEndpoint rates are USD per token. Divide displayed USD-per-million prices by 1,000,000. Published request fees count once per request. With no writes, W5 = W1 = 0. Write/TTL, conditional discount and separately priced reasoning adapters remain gated: this equation does not make those totals complete.
A hypothetical worked example
Using the saved October 4, 2026 Claude Opus 5.5 / Anthropic capture ($4/M input, $20/M output), a hypothetical uncached request gives:
10,000 × 0.000004 = 0.04 USD input
2,000 × 0.000020 = 0.04 USD output
= 0.08 USD per request
10,000 identical requests × 0.08 = 800 USDCounts are fabricated to illustrate math, not observed usage or a validated bill. The public rate source may now differ from that saved capture.
Reported charges
Covered reported charge = sum of original valid charge fields
for unique, nonconflicting request IDsCurrent prices never replace missing historic charges. Money uses integer decimal coefficients, not binary floating-point sums or per-request cent rounding. Tiny nonzero amounts stay visible, with exact strings in evidence. Reasoning included in output is not added again.
Report a formula issue
Use Feedback. Identify the category, expected formula and public source. No log or receipt is attached automatically. Keep prompts and keys private. See reasoning accounting and our evidence policy.
Sources and further reading
These explanations describe Examenos's data and display policy. Calculator examples are explicitly hypothetical. Inspect dated endpoint evidence and the formulas; projections do not authenticate a bill.