How this site works

Automated pipelines fetch provider prices daily, validate them, and rebuild the site. Timestamps show when each dataset was fetched.

Sources

  • LLM list prices: LiteLLM's model price registry (MIT), the de-facto standard machine-readable price file, updated within days of model launches.
  • Marketplace quotes: OpenRouter's public models API (transformed and attributed, per their terms).
  • Cross-check: Helicone's public cost data (Apache 2.0). Sources disagreeing by more than 50% are excluded from cheapest-price verdicts while remaining visible as that provider's listed price, and 30%+ drift is logged for weekly review.
  • GPU rental: live asks from Vast.ai's marketplace API, RunPod's public GraphQL API, and Verda's on-demand and spot list prices.
  • Local hardware: live product price and availability from GMKtec's public Shopify product feed; processor and memory specifications from AMD.

What the numbers mean

  • Token prices are provider list prices per 1M tokens. Batch discounts, caching and volume deals can take real costs well below list.
  • GPU "from" prices are the cheapest live quote per single GPU at fetch time. Marketplace and community tiers are peer hardware: cheaper, fewer guarantees. Secure/on-demand is datacenter capacity.
  • Self-hosting figures are planning estimates from parameter counts and rough throughput heuristics, clearly labeled. They are not benchmarks.

Validation

Every refresh passes automated gates before publishing: sanity bounds on every price, minimum row counts (a broken fetch can't wipe the site), and cross-source drift checks. Failures block the publish and page a human.

Funding

Some outbound provider links are referral links; see the full disclosure. Rankings are ordered by price alone; providers cannot pay to move up a table.

LLM prices 2026-11-22 ยท GPU quotes 2026-11-22