Demand
Leading OpenRouter Apps by Monthly Token Rank
Rank 1 is best. A gap means the app was outside the returned top 100; the API cannot identify the exact lower rank.
Monthly token-usage rank for a fixed July 2026 cohort of six public attributed OpenRouter apps. Gaps mean outside the returned top 100; they cannot distinguish rank 101 from rank 5,000, zero usage, private/hidden status, or unattributed traffic. This is not a whole-market ranking.
Leading OpenRouter Apps by Monthly Token Rank
- Hermes Agent
- Kilo Code
- Claude Code
- OpenClaw
- Cline
- pi
In 2026-08, the fixed cohort's returned ranks were Hermes Agent (#1), Claude Code (#2), Kilo Code (#3), Cline (#5), pi (#6), OpenClaw (#7); lower ranks are better.
Methodology
The source is OpenRouter's official authenticated GET /api/v1/datasets/app-rankings endpoint with sort=popular. The updater makes one request for each explicit inclusive UTC calendar-month window, beginning 2025-01-01, the dataset floor, and ending with the latest completed month. It sets limit=100 and offset=0 instead of accepting the endpoint's default 30-day window or default 50 rows, validates the echoed dates and absolute ranks, and rewrites the CSV in deterministic timestamp and series-ID order.
To keep the ranking chart legible, the series are fixed to the six public apps ranked 1 through 6 in July 2026, the latest complete month at initial publication: Hermes Agent, Kilo Code, Claude Code, OpenClaw, Cline, and pi. Selection uses the endpoint's stable numeric app_id; labels use the returned app names. Every monthly request retrieves only the top 100. A gap therefore means the selected app was outside that returned top 100; the response cannot distinguish rank 101 from rank 5,000, zero usage, a private or hidden app, or traffic missing public attribution. Gaps are not imputed.
For each requested calendar window, popular rank orders public apps by total token volume, defined by OpenRouter as prompt_tokens plus completion_tokens across models. Values are the endpoint's absolute ranks after no category or subcategory filter. Rank 1 is the highest token total, so lower numerical values are better. The chart stores rank rather than token counts and makes no interpolation between monthly observations.
This is an OpenRouter-specific attributed-traffic ranking, not an all-provider AI app market-share ranking. Hidden and private apps are excluded. Apps without the required HTTP-Referer attribution header do not appear as public app entries; X-OpenRouter-Title can set the display name but does not create an entry by itself. OpenRouter merges traffic from related aliases into the canonical visible app, so one displayed app may include multiple aliases. Token totals also reflect each model's tokenizer and therefore are a platform usage measure rather than a standardized amount of text or work.
The package covers 20 complete months from January 2025 through 2026-08 and contains 63 observed app-month rows: 6 for Hermes Agent, 17 for Kilo Code, 8 for Claude Code, 7 for OpenClaw, 20 for Cline, 5 for pi. Every complete raw API response and its metadata are retained in the dated archive. The updater requires OPENROUTER_API_KEY only in the Authorization header and never writes it to output.
Frequently asked questions
What does an app's monthly rank measure?
It is the app’s absolute position among public attributed OpenRouter apps ordered by prompt plus completion tokens in that exact UTC calendar month. Rank 1 has the highest token total. The value is not a rank recalculated only among the six displayed apps.
Why are some apps missing in earlier months?
The updater requests only the top 100 for every month. A gap means the selected app was outside that returned set, but the response cannot distinguish rank 101 from rank 5,000, zero usage, private or hidden status, or unattributed traffic. The chart does not convert the gap to zero or rank 101 and does not interpolate it.
Why track these six apps, and does that introduce hindsight bias?
The cohort is fixed to the six apps ranked 1–6 in July 2026, the latest complete month at initial publication. This keeps the series stable, but it introduces hindsight bias: earlier periods follow apps known to lead later, not each earlier month’s top six. The chart should not be used to identify every historical leader.
Which OpenRouter apps are excluded from the ranking?
Hidden and private apps are excluded, as is traffic that cannot be represented as a public attributed app. Therefore, the ranking describes the visible public-app population in OpenRouter’s dataset, not all application traffic routed through the platform.
How do app attribution and alias merging affect the results?
Public attribution requires an HTTP-Referer identifying the app URL; X-OpenRouter-Title can set a display name but does not create an entry by itself. OpenRouter merges related aliases into a canonical visible app, so one ranked app may combine traffic attributed through multiple aliases.
How are the monthly ranking windows defined?
Every observation uses one explicit, complete UTC calendar month, from its first day through its last day. The chart begins in January 2025, the dataset floor, and excludes the current partial month. Ranks are not trailing 30-day values, and missing months are not interpolated.
Why do these rankings matter to the AI ecosystem?
They show which public applications are generating relatively more model-token demand through OpenRouter and how leadership changes over time. That helps track application-layer competition and workload concentration within this routing platform, while leaving direct-provider and other-platform activity unmeasured.
Does rank measure an app's revenue, users, or market share?
No. Rank reflects token volume in public attributed OpenRouter traffic only. It does not reveal token gaps between apps, revenue, paying customers, active users, request counts, retention, profitability, or usage outside OpenRouter, so it is not a business-size or whole-market-share ranking.
Does this chart rank apps across the whole AI market?
No. It covers only public attributed usage routed through OpenRouter and ranks by OpenRouter token totals. It excludes traffic on other providers and non-public or unattributed OpenRouter traffic, so it must not be interpreted as global app usage or market share.