# OpenRouter public-series method

Label: OpenRouter public text-token telemetry

All models within published scope: True

Grouping: Nine named most-used models plus Others each week; model membership changes.

Excludes: Private user/app traffic, private models/endpoints and zero-data-retention traffic; not all platform or global AI demand.

Token types: Published count is prompt+completion tokens using upstream native tokenizers; not normalized compute.

Paid free: Free model variants included and separately named; no complete paid/free series because Others aggregates variants.

Cache: Usage schema places cached tokens in prompt-token details. Chart has no cache breakdown; current cached share of this twelve-week series is unknown.

Period: Monday labels and seven-day spacing; ends reconstructed as following Sunday. Daily buckets documented UTC; weekly boundary timezone inferred from those daily buckets, not separately stated by tooltip.

Unfinished week excluded: 2026-10-19 through2026-10-25

Precision: Displayed compact totals, not exact export. Treat derived percentages as approximate; do not substitute third-party extra digits.

wow: 100*(W[t]/W[t-1]-1)

four_week_endpoint: 100*(W[t]/W[t-4]-1)

four_week_block: 100*(sum(current4)/sum(previous4)-1)

peak: 100*(latest/max(observed_complete_weeks)-1)

Source: https://openrouter.ai/rankings

Observed public-token growth decelerated and latest completed week declined; latest four-week total still grew strongly. Persistent contraction and market-wide demand decline are not established.