Hello, I'm Jonathan R. Reed

I built AI Stats because choosing a model kept turning into fifteen tabs: pricing pages, provider docs, benchmark tables, route catalogs, and speed claims that all used different formats. The project is inspired by Theo Browne's model prices dashboard, with extra catalog and benchmark context for practical model selection.

Pricing, Artificial Analysis benchmark scores, speed, and latency metrics are sourced from Artificial Analysis via their free API.

Epoch benchmark runs are sourced from Epoch AI's AI Benchmarking Hub and are used under Creative Commons Attribution licensing, with Apache 2.0 licensing for Aider Polyglot and Terminal Bench derived subsets.

Model availability, routing metadata, provider directory data, usage rankings, embedding models, modalities, context length, and route pricing are also sourced from OpenRouter public catalog data, with additional public catalog signals from Hugging Face and LiteLLM for popularity, tags, pricing, context, and capability checks. Public catalog usage rankings and route metadata are treated as routing or availability signals, not benchmark scores.

The goal is to make practical model selection easier: compare the models that are cheap enough to run, fast enough for a product, and strong enough on reasoning, coding, math, and knowledge benchmarks. AI Stats keeps those tradeoffs visible in one place instead of forcing teams to jump between provider docs, pricing pages, and benchmark tables.

The interface uses NebulaSans for its display typography. The project is intentionally small, public, and source aware so the data can be checked, corrected, and improved without hiding the underlying assumptions behind a black box ranking.

AI Stats is public and noncommercial. I use it to learn, compare model tradeoffs, and keep my own model choices honest.