Free$50 of credit when you sign up

Every frontier model.
A tenth of the price.

Claude, GPT, Gemini, Grok, DeepSeek and every other top model behind one endpoint that speaks both OpenAI and Anthropic. Paste a base URL into Cursor, Claude Code or Codex and pay per token: prepaid, capped per key, 90% below list.

Get API key Browse all models
Base URL https://api.tokenlowcost.com/v1
40+models
15labs
−90%vs list price
2Mmax context

What requests cost

Simulated requests, priced at today's rates

Example
RequestTokens in → outList priceYou pay
You'd pay$0
Same requests at list price$0
You'd save$0
Works with the tools you use
  • Cursor
  • VS Code
  • Claude Code
  • Codex
  • Cline
  • Roo Code
  • Continue
  • OpenAI SDK
  • Anthropic SDK
Models

Ranked. Best models first.

The catalog is ordered by capability, strongest first, so the top of the list is the best model money can buy — now at a tenth of its list price.

All models
USD per 1M tokens · list price struck through
#ModelTierContextInputOutputYou save
Loading models…
Compare every model
Labs

15 labs. One account.

Anthropic, OpenAI, Google, xAI, DeepSeek, Qwen and more — one balance, one bill, one base URL. Make a key for each model you want and switch by switching keys.

Product

Built like infrastructure.
Priced like a launch promo.

One endpoint for every protocol your tools speak, with the controls you would expect from a real API provider.

One endpoint. Every protocol.

The format is detected from the path — OpenAI Chat Completions, Responses or Anthropic Messages — so the same base URL works in every tool. The /v1 prefix is optional.

POST/v1/chat/completionsCursor · VS Code · SDKs
POST/v1/messagesClaude Code
POST/v1/responsesCodex
GET/v1/modelsyour key's model

A budget and an expiry on every key.

One key per project, teammate or agent. Cap what it can spend, give it a lifetime, revoke it in one click — enforced on every request.

Side projectclaude-opus-5.5
$3.10 of $5.00expires in 27 days
CI · nightly evalsgpt-6-sol
$17.60 of $20.00near limit
Example

Every request itemized.

Input, cached input, output and the exact charge — per request, per key, per day.

modelclaude-opus-5.5
input48,210 tok
from cache41,600 tok
output1,388 tok
charged$0.00625
Example

Cache hits at cache rates.

Coding tools resend the same context every turn. Whatever the provider serves from its cache is billed at the cached rate.

fresh input
$0.40
cached input
$0.02
Opus 5.5 · USD / 1M tokens

The model you pick is the model you get.

No silent downgrades, no cheaper substitutes. Each response names the model that actually answered.

key modelclaude-opus-5.5
served byanthropic/claude-opus-5.5match
billed asclaude-opus-5.5match
Example

Spend per day, split by model.

Hover any day in the dashboard to see exactly which models the money went to. Every request is listed with its tokens and cost.

Prepaid. No surprises.

Top up from $5 in crypto and spend it per token. No seats, no subscription, no monthly minimum — and no overdraft: at zero, requests stop.

Balance$48.20
Spent this month$11.84
At list price$118.40
Example
Calculator

What would your month cost?

Pick a model and drag the sliders. The numbers use live rates, including the cached-input rate.

Setup

Two settings. Keep your tools.

  1. 1

    Create an account and a key

    Pick a model. You can also set a spending limit and an expiry date. The key is shown only once, so copy it right away.

  2. 2

    Set up your tool

    Every tool uses the same base URL, https://api.tokenlowcost.com/v1. Claude Code and Codex need one pasted line. In Cursor and VS Code, you paste your key in the settings.

  3. 3

    Keep working

    Your tools work as before. The dashboard shows every request and what it cost.


          

        

        
Step-by-step guides for every tool are in the documentation.

      

    


    
    

      

        

          FAQ
          
Straight answers.

          
Setup for every tool is in the docs. Current rates for every model are on the pricing page.

        

        

          

            How can it be this cheap?
            
We're in beta and subsidizing usage to grow: you pay a tenth of each model's list price and we cover the rest. Rates will go up after the beta, and you'll get notice before any change.

          

          

            Is there free credit to start?
            
Yes. Every new account gets $50 on its balance right after sign-up, with no card or payment. It's one per person: an account made from an email or network that already got it starts at $0, and extra accounts made for credit get banned. The rules have the details.

          

          

            Do I really get the model I choose?
            
Yes. Each key is pinned to the model you picked when you created it. That model answers every request made with the key, it's the model you're billed for, and the response names the model that replied. We never swap in a cheaper one.

          

          

            Why is each key tied to one model?
            
Because tools send their own model names — Cursor, Codex and others often use internal identifiers. Pinning the model to the key means that whatever your tool sends, you get exactly what you picked. Need several models? Create several keys.

          

          

            Are cached tokens cheaper?
            
Yes. When the provider serves part of your prompt from its cache, those tokens are billed at the model's cached-input rate — also a tenth of list. Coding agents resend the same context every turn, so on long sessions this is often most of your input.

          

          

            How does billing work?
            
Prepaid. Each request is priced per token and deducted from your balance, down to a millionth of a dollar. Set a budget and an expiry per key. When a key reaches its budget or your balance reaches zero, requests stop — there's no overdraft.

          

          

            Do you store my prompts?
            
No. We keep only what billing needs: time, key, model, token counts and cost. Prompts and completions are passed to the model provider and aren't stored by us.

          

          

            Which tools are supported?
            
Cursor, VS Code, Claude Code, Codex, Cline, Roo Code, Continue, the OpenAI and Anthropic SDKs, and any other tool that works with the OpenAI or Anthropic API. The docs have short setup steps for each.

          

          

            How do I add funds?
            
Open Billing in your dashboard, enter an amount from $5 to $2,000 and pay with USDT, USDC, BTC, ETH or 100+ other coins through NOWPayments. Your balance is credited automatically as soon as the network confirms the payment. If less than the invoice amount arrives, we credit the part that did. A few networks, such as USDT on TRON and Bitcoin, need a larger amount because of their fees; the dashboard shows the current minimum before you pay.

          

        

      

    


    
    

      

        Get started
        
Same models.
A tenth of the bill.

        
Create an account and get $50 of credit to start. Make a key, paste the base URL: setup takes about a minute.

        
        

          Get API key
          
          Read the docs