Tokenify

Tokenify is an OpenAI-compatible reseller. Set TOKENIFY_API_KEY and DEEPSEEK_API_KEY on the server. Its DeepSeek model IDs differ from the official API IDs.

Install ai-lcr, @ai-sdk/deepseek, and @ai-sdk/openai-compatible in your project.

Price-aware routing by time of day

Create one router for each pricing window, then select at the start of every request. The same model can be reused in generateText, streamText, tools, and agents; a stream stays on its chosen route if it crosses a boundary. Tokenify's /v1/models now publishes a separate pricing_offpeak for V4.1 Flash. These are USD prices per million tokens, checked September 29, 2026.

import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
import { createDeepSeek } from "@ai-sdk/deepseek";
import {
  createLCR, createTimeRoutedModel, isDeepSeekPeak,
  DEEPSEEK_PRICES, DEFAULT_PROVIDERS,
} from "ai-lcr";

const tokenify = createOpenAICompatible({
  name: "tokenify",
  baseURL: DEFAULT_PROVIDERS.tokenify.baseURL,
  apiKey: process.env[DEFAULT_PROVIDERS.tokenify.apiKeyEnv],
});
const deepseek = createDeepSeek({ apiKey: process.env.DEEPSEEK_API_KEY });
const direct = deepseek("deepseek-flash"); // Official V4.1 Flash
const reseller = tokenify("deepseek/deepseek-v4.1-flash");

const offPeak = createLCR({
  models: { flash: [
    { model: reseller, label: "tokenify", cost: DEEPSEEK_PRICES.flash.tokenifyOffPeak },
    { model: direct, label: "deepseek", cost: DEEPSEEK_PRICES.flash.officialOffPeak },
  ] },
});
const peak = createLCR({
  models: { flash: [
    { model: reseller, label: "tokenify", cost: DEEPSEEK_PRICES.flash.tokenifyPeak },
    { model: direct, label: "deepseek", cost: DEEPSEEK_PRICES.flash.officialPeak },
  ] },
});

export const flash = createTimeRoutedModel(
  (at) => isDeepSeekPeak(at) ? peak("flash") : offPeak("flash"),
);

Both Flash chains try Tokenify first. Its off-peak fresh input and output are 40% below DeepSeek's off-peak rates; at peak they are 50% below official peak rates. Cache reads tie in both windows. Each chain retains normal ai-lcr failover and recovery. autoSort alone sorts only when a router is created, so use createTimeRoutedModel to attach the correct price to each request.

For V4 Pro, use official deepseek-v4-pro, Tokenify deepseek/deepseek-v4-pro, and DEEPSEEK_PRICES.pro. Tokenify publishes one fixed Pro price: off-peak, put DeepSeek first for its cheaper cache reads; at peak, put Tokenify first for its cheaper fresh input and output. See the side-by-side table.

In Pacific Daylight Time, DeepSeek's peak windows are Sunday–Thursday 18:00–21:00 and 23:00–03:00 the following day; Pacific Standard Time shifts one hour earlier. All other hours are off-peak. The official rule is Monday–Friday 01:00–04:00 and 06:00–10:00 UTC. isDeepSeekPeak uses UTC directly to avoid daylight-saving mistakes. The current rule states no holiday exception.

The cost values are list-price estimates for onCall and onCost. Provider-reported charge, when present, takes precedence; billing at a boundary or future price change can differ from a request-start estimate. Check Tokenify's models API (with your API key; pricing_offpeak is not yet shown in the dashboard's main price cards) and DeepSeek's official prices before deploying a long-lived configuration.

DeepSeek retired V4 Flash. Its legacy deepseek-v4-flash ID now routes to V4.1 Flash on the official API. Tokenify lists V4 Flash separately, so do not assume those two IDs serve the same model version.