# tcp.systems

> Open model prices, seller comparisons, GPU hourly quotes, host pages, lab docs, and an AI repo tape.

The public site is a JavaScript SPA. GET /api/v1/ and GET /api/v1 return the same JSON index. Token job price is 1 million input tokens plus 200 thousand output tokens. The default price sort is chat, reasoning, and code with an output price above zero. Truth order is bill (0), vendor (1), list (2), marketplace (3), context (4). A lower truth wins before a smaller number. Rates under $0.0001 per 1M tokens are sentinels and stay off that rank.

Query parameters on /api/v1/families are case-insensitive: sort, type (chat|reasoning|code|embed|speech|image|video|rerank|all), output=0, truth, q, page, page_size (20|40|80), seller, max_unit, min_context, preference=1, coding=1, open=1. Compare accepts /compare?ids=id,id.

Families show open vs closed when that fact is known. Unknown is omitted, not guessed. Preference and coding are published scores. A missing score means no score; it is not zero, and a value above 1 is not a 0–1 pick rate. GPU quotes on /gpus are USD per hour and stay out of the token ranking. /hosts redirects to /gpus. Embeds, media, and tiers are their own boards.

## Desks

- [Overview](https://dev.radar.tcp.systems/): interactive desk
- [Models explorer](https://dev.radar.tcp.systems/explore): family table with prices, open vs closed when known, and scores
- [Model page](https://dev.radar.tcp.systems/m/{id}): one family. Replace {id} with the family id from /explore or /api/v1/families
- [GPUs](https://dev.radar.tcp.systems/gpus): hourly GPU board. Not mixed into the token ranking
- [Embeds](https://dev.radar.tcp.systems/embeds): embedding prices, not the token job
- [Media](https://dev.radar.tcp.systems/media): per-image, per-second, and per-minute prices
- [Tiers](https://dev.radar.tcp.systems/tiers): plan tiers
- [Scores](https://dev.radar.tcp.systems/scores): preference and coding boards. Missing is no score
- [Compare](https://dev.radar.tcp.systems/compare): side by side families. /compare?ids= keeps the pair
- [Providers](https://dev.radar.tcp.systems/providers): cheapest-job wins. A subscription or flat catalog has no win count
- [Docs](https://dev.radar.tcp.systems/docs): stored lab llms.txt pages
- [Repos](https://dev.radar.tcp.systems/repos): Awesome AI snapshot
- [Sources](https://dev.radar.tcp.systems/sources): fetch health. A 200 with zero parsed prices is not OK
- [Guide](https://dev.radar.tcp.systems/guide): where to start

## API

- [API root](https://dev.radar.tcp.systems/api/v1/): JSON index
- [config](https://dev.radar.tcp.systems/api/v1/config): desk config and score tests
- [families](https://dev.radar.tcp.systems/api/v1/families): model families; one family at /api/v1/families/{id}
- [boards](https://dev.radar.tcp.systems/api/v1/boards/scores): score, GPU, media, tier, and embed boards. Each score board has total and rows
- [gpus](https://dev.radar.tcp.systems/api/v1/gpus): hourly GPU hosts and quotes
- [sources](https://dev.radar.tcp.systems/api/v1/sources): fetch surfaces