Skip to content

Show your work

The numbers should be
easy to challenge.

Every estimate comes from published list prices and the formulas below. If we cannot price something without guessing, we say so and leave it out.

Catalog
2026-08-15
Records
48
Review window
90 days

Core calculation

The formulas

Token rates are stored per million tokens, just as vendors publish them. Currency uses precise decimal arithmetic and is rounded only when displayed.

cost/request = (inputTokens  × inputRate
              + outputTokens × outputRate
              + cachedTokens × cachedRate) / 1,000,000

monthlyRequests = monthlyActiveUsers × requestsPerUserMonth
monthlyAI       = cost/request × monthlyRequests
cost/user       = monthlyAI / monthlyActiveUsers

stackMonthly    = sum(selected tool plans at entered usage)
budgetForAI     = max(0, monthlyBudget - stackMonthly)
affordableTotal = floor(budgetForAI / cost/request)

Cost per request is shown to six decimal places; monthly totals to two. The budget calculator prices tools at the entered user count, then calculates the AI allowance from the remaining budget.

Plan selection

How stack estimates work

For each selected tool, we price every recorded plan and select the lowest-cost plan that can serve the entered usage. Ties use the lower base fee and then the plan ID.

  • Blocked: the plan has a hard cap and becomes unavailable above it.
  • Linear: each started vendor billing unit is rounded up and charged.
  • Graduated: usage moves through cumulative brackets at each published rate.
  • Included credit: offsets usage charges, never the plan's base fee.
  • First limit: keeps editable per-user ratios constant as user count grows.
  • Unmodeled: adds nothing to the total and always explains why.

Budget estimates hold tool usage at the entered user count. Free-plan estimates use the current stack total as the editable non-AI baseline.

Scope

What is included—and what is not

Included

  • Published input and output token rates.
  • Cached-input rates where the vendor publishes one.
  • The same entered workload across every comparison.

Excluded

  • Taxes, contracts, volume discounts, and enterprise pricing.
  • Batch discounts and regional premiums.
  • Tool calls, web search, code execution, and storage charges.
  • Image, audio, video, cache-write, and context-tier pricing.

An excluded pricing dimension is shown beside the result. It is never silently estimated.

Source policy

How pricing data is maintained

  • Model records are read by a person from the vendor's primary pricing documentation.
  • Tool records are fetched from the vendor's published page and marked pending a human spot-check.
  • Every record must have a source URL, verification date, and reviewer.
  • Records older than 90 days show a visible out-of-date warning.
  • Catalog changes go through human review and automated checks before publication.
Current version2026-08-15
Oldest verification2026-08-15
Inspect every record →

Cost only

Comparisons

Comparisons keep usage identical and sort by calculated price. Cheaper never means better, more capable, or more suitable. If a record cannot be priced for the entered usage, we show the reason instead of silently dropping it.

Independence

Sponsors

Sponsors are always labeled and cannot change prices, results, comparison order, or catalog coverage. Their configuration has no access to calculator inputs or results.

Read the sponsorship policy →

Found something wrong?

Send us the source.

Go to Pricing data and submit the catalog item, corrected value, and vendor source URL. A person reviews every correction before public data changes.

Go to Pricing data