LLM API glossary: token, context window, MTok, routing and more
A short, plain-English glossary of the terms that come up when you call an LLM API — token, context window, price per million tokens (MTok), routing, OpenAI-compatible, Anthropic-compatible, x402, discount and healthy provider — with how each one applies when you use UberLLM.
A plain-English reference for the words that show up on pricing boards, in docs and in gateway routing. Each entry says what the term means and how it applies when you call models through UberLLM.
Token
The unit an LLM reads and writes. Text is split into tokens — roughly a word or a word-piece — before the model processes it, and both your prompt (input tokens) and the model’s reply (output tokens) are counted. LLM APIs bill per token, which is why a tighter prompt and a smaller max_tokens cost less. See live per-token prices on the pricing board.
Context window
The maximum number of tokens a model can consider at once — your prompt plus the reply must fit inside it. A larger context window lets you pass more documents or a longer conversation, but every token in the window is still an input token you pay for. Each model’s context length is listed on /models.
$/1M tokens (MTok)
How LLM prices are quoted: dollars per one million tokens, written "/1M" or "per MTok". Input and output are usually priced separately (output is typically dearer). To estimate a call’s cost, multiply its token count by the per-MTok rate and divide by a million. On UberLLM the quoted per-MTok price already includes the 10% fee, so it is the price you pay.
Routing
Choosing which provider serves a given request. A gateway like UberLLM routes each request to the cheapest healthy seller for the model you named, per request — see how routing works. Without a gateway, "routing" is you hard-coding one provider.
OpenAI-compatible
Describes an endpoint that accepts the same requests and returns the same responses as the OpenAI API, so the OpenAI SDK works against it by changing only the base URL and key — full explanation in what is an OpenAI-compatible API endpoint. UberLLM is OpenAI-compatible.
Anthropic-compatible
The same idea for Anthropic’s API: an endpoint that speaks the /v1/messages request and response format, so the Anthropic SDK (and tools built on it, like Claude Code) work against it by changing the base URL and key. UberLLM exposes an Anthropic-compatible /v1/messages endpoint — Anthropic SDK doc.
x402
An open, HTTP-native way to pay for a request using the standard HTTP 402 "Payment Required" status code: the server answers a call with a 402 and payment details, the client pays and retries, and the request is served — no account or subscription needed. It is designed for machine-to-machine and pay-per-request billing.
Discount
On a marketplace, how far a model’s live price sits below its list (direct) price. A seller offering spare capacity lists it below the official rate; the discount is that gap, and it moves as offers change. The live saving per model is shown on the pricing board.
Healthy provider
A seller that is currently serving requests successfully — not erroring, not rate-limiting, not probing unhealthy. Routing only ever sends you to a healthy provider, which is what keeps "cheapest" from meaning "broken". You can inspect provider reliability on the uptime page.