About & methodology
LLM API Price Tracker lists the standard pay-as-you-go API price of language models, per 1 million tokens, from model makers and from the aggregators, inference hosts and clouds that resell them.
How prices are collected
- Every number is read from the provider's official pricing page. If a value isn't on the page, we leave it blank rather than guess.
- The main figure is always the provider's standard price. Where a provider runs a temporary discount and states the regular price, we show the regular price and store the promo price and its end date separately. Those models carry a Promo until … badge.
- Batch, flex, off-peak and promotional discounts are shown on model pages but not used in the table or calculator.
- Models marked variable pricing cost more or less depending on prompt length (long-context tiers), caching (cache writes/reads), or time of day (e.g. DeepSeek peak/off-peak). For these we show the short-context / peak tier, and list the other tiers on the model page.
- "Blended" is a weighted average of input and output price. The default assumes 3 input tokens per output token, (3 × input + output) / 4. You can switch the home table to 1:1 or 10:1. It's only a sorting aid; for real estimates use the cost calculator with your own token counts.
- A script re-fetches the source pages and flags differences for review. Last full verification: 2026-10-15.
Provider types, regions and EU residency
- First-party: the model maker's own API. Aggregator: one API routed to many upstreams (OpenRouter; prices from its public models API, which can vary by upstream). Inference host: serves open-weight and partner models (Fireworks, DeepInfra, Groq, Together AI). Cloud: hyperscaler or regional cloud platform (Amazon Bedrock, Google Vertex AI, Scaleway, IONOS).
- Regions and EU data residency are filled only when the cited source page states them; otherwise they read "unknown". "Partial" means some models or regions qualify, so see each model. Model-level values (for example Bedrock EU-region prices) override the provider value.
- Headquarters country comes from public company information.
- Most prices are in USD. Scaleway, CamoCopy and GreenPT publish prices in EUR (shown with €) and Infomaniak in CHF; we keep them as published and do not convert them, so cross-currency sorting is only approximate.
- See EU LLM providers and aggregators & hosts.
How to use this data
How to use this data. Price per token says nothing about quality, latency, rate limits or reliability. Shortlist a few candidates, test them on your own tasks and measure the total cost per task, then confirm the current rate and exact model ID with the provider before switching.
Sources
- https://platform.openai.com/docs/pricing
- https://docs.anthropic.com/en/docs/about-claude/pricing
- https://ai.google.dev/gemini-api/docs/pricing
- https://api-docs.deepseek.com/quick_start/pricing
- https://docs.x.ai/docs/models
- https://console.groq.com/docs/models
- https://www.together.ai/pricing
- https://openrouter.ai/api/v1/models
- https://docs.fireworks.ai/serverless/pricing
- https://deepinfra.com/pricing
- https://aws.amazon.com/bedrock/pricing/
- https://cloud.google.com/vertex-ai/generative-ai/pricing
- https://www.scaleway.com/en/pricing/model-as-a-service/
- https://cloud.ionos.com/managed/ai-model-hub
- https://api-v1.camocopy.com/docs
- https://router.requesty.ai/v1/models
- https://docs.greenpt.ai/pricing
- https://www.infomaniak.com/en/hosting/ai-services/prices
- https://docs.mistral.ai/inference/pricing
- https://cloud.sambanova.ai/plans/pricing
- https://tokenfactory.nebius.com/api/public/models_info
- https://novita.ai/model-api/pricing
Not yet covered
- Groq (pricing page): Returned marketing copy only; prices taken from console.groq.com/docs/models instead. Llama models listed as 'Contact Sales'.
- OpenAI (openai.com/api/pricing): Shows ChatGPT business plans; API prices taken from platform.openai.com/docs/pricing.
- Azure OpenAI: Price cells render as '$-' (filled client-side); no numbers readable.
- OVHcloud AI Endpoints: Catalog shows prices without model names or units; could not map prices to models.
- STACKIT AI Model Serving: Pricing is in the STACKIT calculator (not in page HTML); model docs list no prices.
- Aleph Alpha: 404; no public per-token price page found.
- IONOS docs pricing: 404; prices taken from cloud.ionos.com/managed/ai-model-hub instead.
Checked, no public API pricing
We checked these and did not add them, because they have no real developer API or no readable published API prices. Chat-app subscriptions are not API prices.
- Proton Lumo (https://proton.me/lumo/pricing, checked 2026-10-14): Chat app only (Free / Lumo Plus / Business subscriptions). No official public developer API or API price list; an API is only mentioned as in development.
- xPrivo (https://www.xprivo.com/, checked 2026-10-14): Chat app only (free tier / PRO subscription), open-source self-hostable, bring-your-own API keys. No API of its own with published prices.
- Cerebras Inference (https://www.cerebras.ai/pricing, checked 2026-10-14): Real developer API exists, but the public pricing page HTML has no readable per-model $/1M token table (JS-rendered Developer Tier Pricing). Models like gpt-oss-120b are documented elsewhere without extractable list prices on the pricing page.
- Hyperbolic (https://www.hyperbolic.ai/docs/inference/performance-limits, checked 2026-10-14): Serverless inference API and playground retired; docs say old model endpoints no longer work. GPU rental only until a new inference product launches.
Data access
The full dataset is at /api/prices.json (CC BY 4.0, schema v3; field definitions are included in the file under schema). LLM-friendly summary at /llms.txt.
Covering 22 providers. Spot an error? Prices change often, so please tell us.