Models & Makers

Methodology

What the index claims. What it does not.

This page describes the ingest. The snapshot date is the board stamp, not a claim that we ran the evals.

SourceArtificial Analysis ingest

Source

Rankings come from Artificial Analysis. The content-pipeline Worker snapshots their Data API (with a Firecrawl fallback) into content/leaderboard_data.json. The previous file is kept as leaderboard_data_prev.json so weekly movers can show score deltas.

Licenses

License labels are inferred from model and company names when the API does not ship a field: Qwen, Llama, Mistral, Gemma, DeepSeek, and similar families map to Open Weights; closed APIs stay Proprietary. Inference lives in workers/content-pipeline/src/license.ts and scripts/infer_license.py. If a vendor publishes a different grant, the snapshot is wrong until the next refresh.

Non-claims

  • We do not run the evals. AA does.
  • We do not pick a “best” model for your stack. Use case still wins.
  • We do not invent versions or scores absent from the current snapshot.
  • Journal posts are 1/day with randomized UTC slots. Thin comparison scorecards on /compare are noindex.

Comparisons (BOFU)

Long-form comparison pages link back here and to the live board. They are the indexable alternative to thin /compare scorecards (noindex).

Corrections

Wrong license or a stale board: hello@modelsandmakers.com.