Where every number in this viewer comes from. Nothing here is measured by us — we collect, match and display data published by the sources below.
OpenRouter
Model catalog, pricing, and live provider performance.
- Columns
- model, ctx, free / tools / reason, $/in, $/out, added, expires, tok/s, TTFT ms, and the per-provider breakdown when you click a model id.
- How
- Catalog and pricing from the public
/api/v1/modelsAPI. Speed and latency (p50 over a rolling 30-minute window) from the endpoint behind OpenRouter's own provider-performance charts — an undocumented page API, not a supported public one, so it can change without notice. - Terms
- See OpenRouter's terms of service.
Artificial Analysis
Independent model benchmarks. Attribution required.
- Columns
- intel (Intelligence Index), coding (Coding Index), gpqa, agentic (τ²-bench), terminal (Terminal-Bench 4.0).
- How
- One call to the Artificial Analysis API per refresh. Each OpenRouter model is matched by exact name to its highest-Intelligence-Index variant. No match means a blank cell, never a guess.
- Credit
- Benchmark data by Artificial Analysis.
Epoch AI
Open benchmark results, used to fill gaps Artificial Analysis leaves.Open benchmark results.
- Columns
- gpqa where Artificial Analysis has none (marked
E), and scicode (SciCode).ECI (Epoch Capabilities Index, one capability score fitted across 60+ benchmarks) and gpqa (GPQA Diamond). - How
- The public benchmark dataset download, once per refresh. We take each model's best score across reasoning-effort levels. SciCode is a different measure from Artificial Analysis's Coding Index, so it has its own column.
- License
- Creative Commons Attribution 4.0 — free to use with credit.
- Cite
Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource].
Epoch AI, ‘Epoch Capabilities Index’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/eci’ [online resource].
- Changes
- None to the scores. We match model names to OpenRouter ids by exact name and pick the best reasoning-effort setting.
LMArena
Human-preference ratings from blind head-to-head votes.
- Columns
- arena (overall) and arena code (coding prompts only), both style-controlled.
- How
- The latest snapshot of LMArena's public leaderboard dataset on Hugging Face, once per refresh. Arena lists reasoning-effort variants separately; we match each to its OpenRouter model by exact name and show the best rating. Ratings measure which answers people prefer, not whether they are correct.
- License
- Creative Commons Attribution 4.0. Ratings unchanged; we only select, match and display them.
- Source
- lmarena-ai/leaderboard-dataset (
text_style_control), published ….
Derived by us
Labels we add on top of the sources above.
- origin
- western / china / unknown, from a maintained lookup of each model author's home country. Routers, cloaked models and labs we couldn't verify stay
unknownon purpose. - value
- ECI points per dollar, using a 3:1 input:output blend of OpenRouter's per-million-token prices, and a ◆ marking the efficient frontier: no other paid model is both more capable on ECI and cheaper (judged within the selected origin when you filter by one). Computed from Epoch AI and OpenRouter data; list prices only, so it ignores how many tokens a model spends per task.
- value
- Intelligence-index points per dollar, using a 3:1 input:output blend of OpenRouter's per-million-token prices, and a ◆ marking the efficient frontier (no other paid model is both smarter and cheaper — judged within the selected origin when you filter by one). Computed from Artificial Analysis and OpenRouter data; list prices only, so it ignores how many tokens a model spends per task.
- domain, in
- Output type and accepted input types, read from OpenRouter's modality string.
This build
…