Start with provider-controlled sources
Official pricing pages, product documentation, API documentation, provider announcements, and provider-controlled structured endpoints are preferred over third-party summaries.
Trust and methodology
How StackLens verifies pricing and calculates cost
StackLens separates provider-published facts, calculated estimates, and our own decision guidance so you can see what is verified, what is inferred, and what still needs testing.
Core rule: missing or conflicting evidence stays unverified. StackLens does not fill unknown pricing, limits, availability, or product capabilities with guesses.
Evidence workflow
From official source to published field
A page can contain several evidence types. Model identity alone does not confirm pricing, context window, or lifecycle status.
Match evidence to the field
Pricing evidence supports price fields. Specification evidence supports context limits. Availability evidence supports lifecycle status. One source is not assumed to verify unrelated fields.
Preserve uncertainty
Missing values remain unavailable or require verification. Conflicting official evidence blocks automated publication until it can be reviewed.
Keep guidance visibly separate
Qualitative notes use labels such as StackLens assessment. They are decision guidance, not claims made by the provider and not substitutes for workload testing.
How to read source confidence
- High confidenceAn official provider-controlled source directly supports the displayed field and the value has a clear interpretation.
- Medium confidenceAn official source supports the value, but a product transition, tier rule, billing boundary, or documentation ambiguity still requires care.
- Low confidence or review requiredThe value should not be treated as purchase-ready. Verify it independently; automated publication should remain blocked where required evidence is missing.
What “Last checked” means
The date shows when StackLens reviewed the linked evidence. Providers can change prices, quotas, plan names, or terms after that review.
A recent date improves traceability but does not replace checking the official page before committing budget or production traffic.
Calculation boundary
How token cost estimates are calculated
Calculators use the selected pricing profile and the workload values entered on the page. They are planning estimates, not provider invoices.
Input and output token cost
input cost = input tokens / 1,000,000 × input price per million
output cost = output tokens / 1,000,000 × output price per million
monthly estimate = (input cost + output cost) × monthly requests
output cost = output tokens / 1,000,000 × output price per million
monthly estimate = (input cost + output cost) × monthly requests
When a supported cached-input rate is selected, eligible cached tokens use that tracked rate. Unsupported cache assumptions are not silently applied.
Common exclusions
- Provider-specific message wrappers and hidden tokens
- Images, audio, tools, storage, search, or other non-token charges
- Retries, agent loops, failed requests, and human review
- Taxes, negotiated contracts, regional pricing, and volume terms
- Quality, latency, reliability, and switching cost
Decision boundary
What StackLens does not claim
No universal winner
A lower tracked price does not establish better quality, reliability, latency, privacy, or developer experience.
No equal-output assumption
Two models may produce different output lengths, retry rates, and review burden for the same task. Those differences can change real total cost.
No final purchasing quote
Plan limits, taxes, enterprise terms, rate limits, and provider changes can differ from the scenario shown. Verify official terms before purchase.
FAQ
Questions teams ask before choosing
Does a model-list API confirm model pricing?
No. A model-list response can support model identity or availability, but StackLens requires separate official pricing evidence before publishing token rates.
Does the lowest estimated cost identify the best model?
No. Cost estimates do not establish output quality, latency, reliability, retry behavior, regional availability, or workflow fit. Test representative requests before choosing a model.
What does Last checked mean?
It records when StackLens last reviewed the cited source. It is not a guarantee that a provider has made no changes since that date, so official terms should still be checked before purchase.
How can I report outdated pricing?
Use the feedback link on an affected page or email [email protected] with the page path, official source, and correction. Do not include API keys or private prompts.