# AI - Paski.dev

> A curated catalog of 112 AI models compared on 7 benchmarks, with the source on every figure. Nothing here is measured by us: every score links the document its maker or an independent platform published. 278 figures published by vendors, 88 measured by third parties.

## Data

- [Catalog as JSON](https://ai.paski.dev/api/catalog.json): every model, benchmark and score with its source_url. Free to use with attribution.
- [Log feed](https://ai.paski.dev/log/rss.xml): RSS for the writing.
- [Sitemap](https://ai.paski.dev/sitemap.xml)

## Pages

- [Comparison table](https://ai.paski.dev/): all models against all benchmarks, filterable by organisation and by whether you can self-host.
- [Benchmarks](https://ai.paski.dev/benchmarks/): what each benchmark measures, how it is scored, and how to run it yourself.
- [Value](https://ai.paski.dev/value/): cost against standing. 112 models plotted; most are beaten outright by something cheaper and no worse.
- [Token cost](https://ai.paski.dev/tokens/): estimate what a prompt costs across every model with a published price.
- [Marketplace](https://ai.paski.dev/marketplace/): a curated index of plugins and skills, crediting their authors.
- [Log](https://ai.paski.dev/log/): findings from the catalog, written from the data.

## Method

- Figures are never generated here. Each one cites the vendor document or independent platform that published it.
- Vendor figures and independently measured figures are kept apart and shown in different colours, because they are different claims.
- Where a source states the reasoning effort a run used, the figure carries it.
- An empty cell means unpublished, not zero.
- Prices are vendor list rates as of the date shown, excluding batch, cached-input and priority tiers.