TokenmaxxingAI token costsAgents online
GuidesLabsEventsLeaderboardToolsBriefing
OpenRouter Rankings

OpenRouter rankings: the most-used AI models

Which AI models are developers actually routing traffic through? OpenRouter publishes daily token volumes for every model on its platform. This page surfaces those rankings with honest data labels and plain-English context — no benchmark scores, no vendor claims.

Source: OpenRouter public rankingsLatest complete day OpenRouter publishes (2026-06-08)
Source-labeled data

Token volume by model, ranked

Each row is one model's token throughput on the latest complete day OpenRouter has published. The 30-day column is a run-rate estimate from that single day — not a rolling total and not a global usage figure. OpenRouter publishes these numbers with a lag of several days; the source date in the table is the day the data actually covers.

OpenRouter rankings: liveToken totals come from OpenRouter's public rankings data, which publishes complete days with a few days' delay — the board shows the latest complete day. The 30-day column is a run-rate estimate from that day, not a global market total.Last checked 6/15/2026
Public OpenRouter model usage. Daily values are source-day totals; 30-day values are run-rate estimates, not global model totals.
RankModelSource dayTokens/day30-day run-rate
#1Deepseek V4 FlashDeepseek2026-06-084.6T139.3T
#2Minimax M3Minimax2026-06-084.5T136.1T
#3Hy3 PreviewTencent2026-06-084.3T128.4T
#4Mimo V2.5Xiaomi2026-06-083.7T112T
#5Owl AlphaOpenrouter2026-06-082.6T77T
#6Claude 4.7 OpusAnthropic2026-06-082.5T74.7T
#7Claude 4.6 SonnetAnthropic2026-06-082.2T65.4T
#8Deepseek V4 ProDeepseek2026-06-082.1T62.2T
#9Claude 4.8 OpusAnthropic2026-06-081.3T38.7T
#10Deepseek V3.2Deepseek2026-06-081.2T35.7T
#11Gemini 3 Flash PreviewGoogle2026-06-081T30.4T
#12Nemotron 3 Ultra 550b A55bNvidia2026-06-08822.8B24.7T
#2
Minimax M3Minimax
Source day
2026-06-08
Day
4.5T
30d run-rate
136.1T
#3
Source day
2026-06-08
Day
4.3T
30d run-rate
128.4T
#4
Mimo V2.5Xiaomi
Source day
2026-06-08
Day
3.7T
30d run-rate
112T
#5
Owl AlphaOpenrouter
Source day
2026-06-08
Day
2.6T
30d run-rate
77T
#10
Source day
2026-06-08
Day
1.2T
30d run-rate
35.7T
Context

What the OpenRouter rankings actually show

These rankings are a measure of revealed preference — what developers paid to route through OpenRouter on one specific day. That is a different signal from benchmark scores or vendor announcements.

Revealed preference, not capability tests

Benchmark leaderboards measure performance on standardized tasks. OpenRouter rankings measure what developers actually chose to send traffic through — at real prices, for real workloads. A model can top the OpenRouter rankings because it is cheap and fast, even if it is not the highest scorer on any capability benchmark.

The data is a few days behind

OpenRouter publishes complete-day totals with a lag — typically three to seven days. The table shows the latest date with a full row set. If a new model launched this week, it may not appear yet. The source date in the table header is the day the data covers, not today.

Token share vs request count

The rankings are ordered by token volume, not request count. A model that handles long-context document work or multi-step agent tasks will generate more tokens per request than a model answering short chat messages — so token volume and request count can tell different stories about the same day.

How to read them

Token share, run-rate, and what the numbers mean

The table has three data columns: source day, tokens per day on that source day, and an estimated 30-day run-rate. Here is what each one tells you — and what it does not.

Source day

The specific calendar date the token figures cover. OpenRouter does not publish a live running total; it publishes complete days. If the source date is several days behind today, that reflects the normal publish lag — not a data quality problem.

Tokens per day

The total prompt plus completion tokens that model processed through OpenRouter on the source day. A cheap model can rank highly here because low price attracts high volume. A model with high per-token pricing may rank lower even if it handles high-value work, because developers send fewer tokens through expensive routes.

30-day run-rate

The source-day total multiplied by 30. This is a rough estimate of what monthly volume would look like if that one day were representative. It is labeled as a run-rate because the underlying data is a single day, and a single day can be unusually high or low. Do not read it as a confirmed 30-day total.

Why it matters

Connecting rankings to routing cost

If you are routing production AI traffic, the OpenRouter rankings are a useful sanity check on your model choices — not a prescription, but a signal about what similar developers are paying for.

High volume often means low price

Models near the top of the token-volume rankings are often there because they are priced to attract volume. If a model you have not considered is consistently in the top five, the most likely reason is that other developers found it cheap enough to route at scale. Worth checking the pricing before dismissing it.

Rank movement is the useful signal

A model moving from rank 8 to rank 2 week-over-week is more useful information than its absolute rank on any single day. It suggests a pricing change, a capability update, or a shift in what developers are building. The full leaderboard tracks these movements alongside cost data.

Rankings are not output quality

A model can rank first on token volume and still be the wrong choice for your workload. Pair any routing decision with your own quality evaluation: accepted output rate, error frequency, and latency at your usage pattern — not just what the rankings say.

Full leaderboard with pricingHow to read OpenRouter rankings
Questions

Frequently asked questions

What is OpenRouter?

OpenRouter is a routing service that lets developers send AI requests to many different models through a single API. Instead of maintaining separate integrations for Anthropic, OpenAI, Google, Meta, and others, developers send requests to OpenRouter and choose which underlying model handles each one. OpenRouter publishes public usage data showing how many tokens each model processes — which is where the rankings on this page come from.

How often do the OpenRouter rankings update?

OpenRouter publishes complete-day usage figures with a lag of several days — typically three to seven days behind the current date. The table on this page shows the latest complete day that OpenRouter has published with a full row set. The 30-day column is a run-rate estimate from that single day, not a rolling total. The source date is shown in the table so you always know exactly which day you are looking at.

Which model is most used on OpenRouter?

The rankings table on this page shows the current answer from OpenRouter's published data. The top-ranked model changes as providers release new versions, adjust prices, and developers switch routing targets. Check the table above for the current source-day leader — it is pulled from OpenRouter's public data at build time and labeled with the exact date.

Are OpenRouter rankings the same as AI benchmarks?

No. Benchmarks test model capability on standardized tasks. OpenRouter rankings measure revealed preference: which models developers actually route real traffic through, on that source day, at whatever price those models charge. A model can top the OpenRouter rankings because it is cheap, fast, or reliable — not because it scores highest on any benchmark. The two signals are useful for different questions.

Sources

Where this data comes from

The rankings on this page come directly from OpenRouter's public data. No adjustments, no estimates beyond the labeled run-rate column. The source-day leader as of 2026-06-08 is Deepseek V4 Flash.

OpenRouter rankings

OpenRouter publishes model usage data on its rankings page and via a public API endpoint. This is the primary source for the table above. The data shows token throughput per model per day, with a multi-day publish lag.

openrouter.ai/rankings

Tokenmaxxing leaderboard

The full leaderboard adds pricing, context windows, open-model momentum, and company spend data alongside the same OpenRouter usage rankings. Use it when you need more than the usage table alone.

Open the leaderboard

Guide: how to read these rankings

A plain-English walkthrough of how OpenRouter token rankings work, what publish lag means, why token volume and request count differ, and how to use the data without over-reading it.

Read the guide