# Cheap GPUs

> FastAPI service that aggregates LLM model pricing and GPU rental rates across multiple providers, and serves a cheapest-route-finder UI.

Cheap GPUs tracks per-token LLM inference pricing across eight providers (OpenRouter, Chutes, Antseed, DeepInfra, Venice, CheapTokens, Novita, SayGM), per-hour GPU rental pricing across eleven providers (vast.ai, Akash Network, RunPod, Lium, Targon, Clore.ai, Vultr, SaladCloud, Scaleway, Lambda Labs, DigitalOcean), and market-implied forward GPU price curves from prediction-market platforms (Kalshi and Polymarket), persisting current snapshots and historical price series to Postgres. Individual GPU providers can be disabled via `GPU_PROVIDER_EXCLUDE` without a code change — see `/status` or `/gpu-status` for which are currently active.

## Pages

- [GPU rentals](/gpu): ranked current cheapest GPU rental offers per GPU family, price history sparkline, and a price heatmap across GPU families over time. The site default.
- [Route finder](/models): cheapest LLM provider for a given model, given input/output token counts.
- [Prediction markets](/markets): market-implied forward price curves per GPU family, derived from prediction-market contracts (Kalshi's weekly GPU-compute-price series and Polymarket's recurring end-of-period price-bracket events) rather than current rentable offers.
- [Model browser](/dashboard): sortable table of every tracked LLM model across all providers (legacy view, superseded by the route finder).

## API

- [Aggregated LLM models](/api/models): every tracked LLM model with normalized pricing in USD per 1M tokens.
- [Sync status](/status): sync health for the LLM, GPU, and prediction-market pipelines, including per-provider status and total counts.
- [Aggregated GPU rates](/gpu-rates): current GPU rental listings across all GPU providers, in USD per hour.
- [GPU sync status](/gpu-status): sync health for the GPU pipeline specifically.
- [GPU price matrix](/gpu-history-matrix): cheapest observed price per GPU family per day across all providers, backing the price heatmap.
- [Aggregated market curves](/market-curves): current forward-curve points (one per provider/GPU family/future expiry date), in USD per hour.
- [Market sync status](/market-status): sync health for the prediction-market pipeline specifically.

Three additional endpoints take path parameters and aren't linkable directly: `GET /history/{provider}/{model_id}` (LLM price history for one model), `GET /gpu-history/{provider}/{gpu_family}` (GPU rental price history for one family), and `GET /market-curve-history/{provider}/{gpu_family}` (how a family's forward curve has shifted over time) — all three accept a `days` query parameter.