Capability Scores & Model Ranking
The SpiderGate catalog tracks 1,216 models across 16 providers and ranks them by what each one is actually good at. An agent keeps the facts current; the coverage is published as a fraction, not rounded up.
Rank by capability, not by benchmark soup
221 of the 1,216 models carry a score in each of nine capability categories. The score is a within-category percentile drawn from 454 benchmark-to-category signals over 4,821 published benchmark results, sourced from LLM-Stats and DesignArena.
Nine capability categories
Coding, Design/Frontend, Reasoning, Math, Vision, Research, Tool Use, Chat and Translation. Filter the catalog by one and it reorders best-first.
A percentile, not an average
An 82 on GPQA and an 82 on SWE-Bench are not the same 82. Each score ranks a model against its peers in that category instead.
An agent keeps the catalog current
Keeping 1,216 model records accurate by hand is nobody's job. A daily enrichment agent pulls the facts from public sources and stamps every field it writes with where the value came from and when.
Runs every day
Specs and pricing from OpenRouter, benchmark results from LLM-Stats, and model lineage from Wikidata.
Provenance on every field
Each value carries the source it came from and the moment it was fetched, so a number can always be traced back.
The coverage is stated, not rounded up
221 of 1,216 models are scored, about 18 percent. Everything else reads not scored yet rather than ranked last, because a model only scores when real benchmarks exist to score it on. Evals are just getting started.
Curated rows are protected
Write editorial copy for a model and that row is marked curated. The daily sync flows around it and never overwrites it.
The benchmarks are visible
Open any score to see the results behind it, with vendor-reported figures flagged as vendor-reported.