Pricing Methodology

Tokencost.live is an independent calculator and research project. Pricing values are collected from official provider pricing pages when possible. If pricing is unavailable, unstable, region-specific or locked behind account terms, calculators use manual input instead of invented rates.

Accepted sources

Official provider pricing pages and docs.
Official cloud marketplace pricing pages.
Open-source pricing datasets only when clearly linked back to a provider source.
Manual user input for plans, credits or private pricing that cannot be verified publicly.

How estimates are calculated

Token cost estimates divide input and output tokens by one million, multiply by provider rates, then apply cache ratio, retry rate and monthly volume. Creator estimates use attempts per usable output. SaaS estimates combine revenue, API cost, hosting, payment fees and fixed costs.

Currency and rounding

Provider data is stored in the currency published by the source. Currency conversion is approximate and should be checked against the user's payment method. Public pages round displayed estimates to cents for readability.

Exclusions

Estimates may exclude taxes, enterprise discounts, minimum monthly commitments, data-residency uplifts, marketplace fees, moderation/tool fees, failed network calls and provider-specific quotas.

Daily pricing refresh process

Tokencost keeps a central pricing file at /data/verifiedPricing.json. The automated sync checks the community-maintained simonw/llm-prices dataset and records a sync report at /data/llmPriceSyncReport.json. Kimi K3, K2.7 Code and K2.6 are currently kept from the official Kimi platform because those prices are published in CNY on the provider site.

The automation does not blindly convert every model into public advice. The research pages use selected representative rows and preserve source links, currency and verification status. This matters because some rows in public datasets can be previews, region-specific, tiered by context length or subject to provider changes. When a row is uncertain, the safer decision is to show the source and ask users to verify before spending money.

Quality control rules

Every indexed pricing page must have a clear purpose beyond repeating a keyword.
Every factual price table should show source, currency and last-verified date.
Manual inputs are used for unclear subscription, credit or regional plans.
Mixed-currency examples are not added together unless an exchange-rate source is stated.
Pages that duplicate another resource are redirected, removed from the sitemap or marked noindex.

Formula review

LLM estimates begin with token volume: input tokens, cached input tokens and output tokens. Agent estimates add turn count, hidden reasoning allowance where relevant, cache ratio and retry overhead. Creator estimates add attempts per finished asset because failed generations are part of the real production cost. SaaS estimates combine revenue, payment fees, fixed costs, hosting, database, storage, API spend, marketing and team cost.

The final number should be read as a planning estimate, not an invoice. Actual bills may include taxes, committed-use discounts, regional data processing, moderation endpoints, tool-use calls, storage, network egress, file search, vector databases, logging and customer support costs.

Related calculators and research

Report an error

Email hello@tokencost.live with the page URL, provider source and correction.

Why this methodology page exists

Tokencost is built for people who need to make a practical money decision before they have perfect data. A developer might be deciding whether a chatbot can fit inside a $19/month subscription. A creator might be checking whether a short-video workflow can survive the cost of regenerated clips, voiceovers and music. A small SaaS team might need to know whether an AI feature should be free, metered, routed to a cheaper model or limited behind a paid plan.

Those decisions are difficult because AI pricing is fragmented. LLMs bill input, output and sometimes cached input. Video and music tools may use subscriptions, credits, region-specific plans or private enterprise rates. Voice tools may charge by characters, minutes or seats. Some providers publish clear API rates while others hide the practical unit cost behind plan limits. Tokencost keeps calculators flexible by allowing manual input when verified pricing is not available.

What makes a useful estimate

A useful estimate includes the whole workflow. For LLM products, that means request count, prompt size, output size, retries, cache ratio, hidden reasoning or agent loops and monthly active use. For creator workflows, it means finished output, rejected attempts, editing tools, revenue assumptions and break-even views. For SaaS businesses, it means revenue, payment fees, infrastructure, marketing, team cost and churn. A single provider price rarely answers the real business question.

How to read Tokencost pages

Use every result as a planning range. The calculators show how cost moves when assumptions change, while the research pages explain why those assumptions matter. The sources and methodology pages show where factual pricing came from and where users should verify. If a provider publishes a newer source, the site can be corrected through the contact email. This structure is intentionally conservative because publishing a transparent caveat is more valuable than showing a precise-looking but unverifiable number.

Editorial promise

Tokencost avoids copyrighted logos, keeps ads separate from calculator logic, removes thin duplicate pages from the sitemap and keeps legal/trust pages accessible without login. The goal is to be a durable reference for AI pricing, token cost calculation, creator cost planning and AI SaaS economics, not a temporary collection of search-keyword pages.