Index = OpenRouter's tokens for the week + OpenCode's tokens for the same week, in trillions. Weeks run Monday to Sunday (UTC) and are dated by the Sunday.
Each week's bar is split by model provider: OpenRouter's tokens per model, by the lab named in the model's ID (deepseek/, z-ai/), and OpenCode's tokens per lab, added together. The six providers with the most tokens over the index's history are named; every other provider, OpenRouter's "Others" (the models outside its weekly top ten), unidentified models and OpenCode's "Other" are counted as Others.
OpenRouter: the weekly token total across every model it routes, as its rankings report it. OpenCode: the daily token totals of its usage data (input, output, reasoning and cached tokens), summed over the week once all seven days are reported.
A week counts in the index only when both platforms report it in full: the week still running is left out. The index starts with the first week both cover.
The platforms revise recent days upward as late usage arrives; each week is recomputed from the latest figures until it stops changing.
Until August 2026 OpenCode described its figures as usage of OpenCode Go, its hosted model plan, and since then as usage of OpenCode. Its daily totals rose about fivefold over the first ten days of August, and earlier days were not restated, so weeks before August likely cover a narrower part of OpenCode than later weeks.
Requests OpenCode sends through OpenRouter are counted by both. They are too few to rank among OpenRouter's top 20 apps, under 0.5 trillion tokens a week against about 400 trillion in the index (November 2026), and are not taken out.
It counts only usage these two platforms publish. Traffic sent straight to model labs' own APIs, and other apps' traffic, is not included: the index is a public meter on demand, not a census of it.
How the open token consumption is built · Indices · The Inference · The Inference