Editorial policy

Who maintains TokenCost

TokenCost is maintained by an independent editorial team focused entirely on LLM API pricing and performance data. We are not a model provider, we do not sell inference, and we have no commercial relationship with any provider whose pricing we display. The site runs as a focused, single-purpose reference rather than a general tech blog, which is why every page you'll find here traces back to a live pricing or benchmark figure.

How pricing data is sourced and verified

Model pricing is compiled automatically from live, third-party provider data (via OpenRouter) on a scheduled daily pipeline — not hand-typed and not manually re-keyed, which is what lets it stay current without drifting out of date. Every figure is normalized to the same unit (USD per 1M tokens) so comparisons are apples-to-apples rather than reconstructed from providers' inconsistent quoting conventions. The full technical breakdown of the pipeline, including exactly which fields are normalized and how cached-input rates are handled, is on our methodology page.

How benchmark data is sourced

We do not run our own model evaluations. Benchmark scores are ingested directly from each benchmark's own published results (LiveBench and Epoch AI's dataset covering SWE-bench Verified and Aider Polyglot) and joined to our pricing data by model name. We add our own commentary and cost-vs-quality analysis on top of those scores, but the underlying numbers are never altered — see benchmarks for the full leaderboards with source links to each.

Original commentary vs. sourced data

It matters which parts of this site are our own analysis and which are third-party data reproduced with attribution. The pricing tables, benchmark scores, and cost-vs-quality charts are sourced data, clearly attributed to their origin. The guides, provider pricing write-ups, and interpretive text throughout the tool pages are original commentary written by the TokenCost editorial team, built from — but not copied from — the underlying data.

Corrections

If a price, model detail, or benchmark figure looks wrong, tell us through the contact page. We check every report against the live source before making a change, and we'd rather be told about an error than have it sit uncorrected. Because pricing is refreshed automatically, most staleness issues self-correct on the next scheduled run — reported errors are typically a mismatch we need to fix in how a figure is parsed or normalized, not a one-off typo.

Independence and no vendor bias

TokenCost is not affiliated with, sponsored by, or endorsed by any model provider, and we do not accept payment for placement, ranking, or favorable treatment of any model. Sort orders, "cheapest" rankings, and benchmark leaderboards are all computed directly from the underlying numbers — there is no manual curation that favors one provider over another. The site is supported by advertising, which is entirely separate from the data and rankings shown.

Update cadence

Pricing and benchmark data run on an automated daily schedule; the "updated" date shown on the pricing page and benchmarks reflects the actual most recent pipeline run, not a fixed or hardcoded date. Editorial guides are dated with their real publish date and are revised when the underlying guidance changes materially.

Questions

Reach us any time through the contact page — for corrections, questions about methodology, or feedback on the guides.