TokenMart — ai-lcr Docs

TokenMart — cheap mainstream catalog. Host is model.service-inference.ai; plain un-prefixed model ids; prompt caching works.

TokenMart (also branded "Inference AI") is a cheap mainstream catalog — Gemini, Claude, GPT, and more at a discount.

import { createOpenAICompatible } from "@ai-sdk/openai-compatible";

const tokenmart = createOpenAICompatible({
  name: "tokenmart",
  baseURL: "https://model.service-inference.ai/v1",
  apiKey: process.env.INFERENCE_API_KEY,
});
// plain ids: tokenmart("gemini-2.5-flash"), tokenmart("claude-sonnet-4-6")

Two things to know:

The host is model.service-inference.ai. api.tokenmart.ai does not resolve — don't use it.
Model ids are plain, un-prefixed (gemini-2.5-flash, not google/…).

Prompt caching works on TokenMart (cache_read fires on repeats), so you can set cacheRead pricing and get accurate cached-token cost.