Why Routerbase

Choosing an LLM shouldn't require reading blog posts. Every provider has a "best model" and every benchmark is designed to make someone look good. I wanted a tool that just showed me the data.

Routerbase pulls model information from OpenRouter — 339 models at last count — and layers on quality scores from LMSYS Arena, speed benchmarks from Artificial Analysis, and weekly popularity data. No editorial. No "best for" recommendations. Just filterable, sortable data.

The filtering is the interesting part. Models differ across so many dimensions that a simple ranked list is useless. You need to filter by what matters for your specific use case: does it support function calling? JSON mode? Vision? What's the price per million tokens? What's the actual measured latency, not the marketing claim?

I built shelf pages for common queries — Best Value, Fastest, Largest Context — because certain comparisons come up over and over. But the real power is the filter combination. Show me models under $1/M tokens that support function calling and have an Arena ELO above 1200. That narrows 339 models to maybe 8, which is a decision I can actually make.

The data pipeline runs on a schedule. OpenRouter model sync, LMSYS quality scores (with fuzzy matching because model names are maddeningly inconsistent across sources), Artificial Analysis speed benchmarks, and popularity rankings. Keeping it current matters because the model landscape changes weekly.

I built this because I use a lot of different models across different projects and I was tired of Googling. OpenRouter's own interface is fine but it doesn't aggregate external quality data. The leaderboards exist separately but don't show pricing. Nobody puts it all in one place, so now I do.

SEO was a deliberate focus. Every model has its own page with structured data, because people search for specific model names. Lighthouse scores 100 across the board. The site is essentially a reference tool and it should behave like one — fast, scannable, and findable.