Question 1

What does the OpenRouter Leaderboard show?

Accepted Answer

Every model listed on OpenRouter, ranked by price per million tokens, by the providers serving it, by first-token latency measured on a paid API, and by a Korean-language quality grade. Prices are shown in USD, KRW and CNY at live rates, and the interface is available in English, Korean and Chinese.

Question 2

How is this different from OpenRouter itself?

Accepted Answer

OpenRouter publishes price, provider and traffic data, which this leaderboard builds on and credits. What it does not publish is whether a model's Korean holds up. That grade is measured here. This site also gathers figures OpenRouter spreads across separate pages into one comparable table, and adds each provider's headquarters country and data-centre locations, which procurement teams ask for.

Question 3

How is the Korean quality grade measured?

Accepted Answer

A judge model grades each candidate on seven axes: register, knowledge of Korean institutions, format compliance, honorific accuracy, figure preservation when summarising, professional terminology, and freedom from translationese. The axis grades are averaged into eight bands from A+++ to F. The judge never sees the model name, it was used only after scoring every known-answer control correctly, Han characters and kana are counted mechanically rather than judged, and a model graded on fewer than five axes receives no grade at all.

Question 4

Is the speed data measured or copied?

Accepted Answer

Measured. Each model was called on a paid API with an identical prompt, recording first-token latency and tokens per second. OpenRouter's own latency percentiles, aggregated over real traffic, are shown alongside so the two can be cross-checked. The single-call measurement and the traffic aggregate are labelled separately rather than merged.

Question 5

How often is the data updated?

Accepted Answer

Daily at 08:00 Korea Standard Time. Prices, rankings and exchange rates refresh automatically, and a snapshot is archived each day so price history accumulates.

Question 6

Is there an API?

Accepted Answer

Yes, and no key is required. /api/models returns every model with unit prices, /api/korean returns the Korean grades including per-axis detail, /api/speed returns the measured latency and throughput, and per-model endpoints return provider-level price, precision, uptime and latency percentiles.

Dashboard

Price distribution

Cheapest by vendor

Best value

Browse models