Open models at a tenth of the list price.
One OpenAI-compatible key. Jobs run on idle GPUs in your region. Every call returns its own receipt.
25 free test calls. No credit card.
Today's tariff
per 1M tokens, in / outThe catalog is being published. Every active model and its price will appear here. See the catalog.
Your OpenAI SDK already works.
Change the base URL and the key. Your prompt formats, streaming logic, tool calls and client libraries stay exactly as they are.
from openai import OpenAI
client = OpenAI(
base_url="https://cheap-tokens.vercel.app/v1", # line one
api_key="ct_live_...", # line two
)
response = client.chat.completions.create(
model="llama-3.1-8b-instruct",
messages=[{"role": "user", "content": "Summarize this in two sentences: ..."}],
)Two lines change. Nothing else does.
A receipt for every call.
Where it ran, first-token time against the promise, tokens, the price, and what the same call costs at list price.
A call that misses its promised time is refunded without you asking.
Sample receipt
Delivered on time- Model
- Llama 3.1 8B Instruct (llama-3.1-8b-instruct)
- Ran on
- eu · RTX 4090 class (host network)
- Tokens
- 214 in · 187 out
- Price
- $0.000007
- Same call at list (Together AI)
- $0.000072
- You saved
- $0.000065 (90%)
When a call is late, a refund line appears on its own: Refunded $0.000007 · late first token. No ticket, no request. The sweep that issues it runs every minute.
Where your prompt goes.
From your app to a host inside your key's region and back. The payload is deleted after completion.
One region per key
A key is pinned to us or eu. Its jobs never leave that region.
Encrypted in transit and at rest
Prompts are AES-256-GCM ciphertext from the gateway to the host.
Nothing touches the host's disk
The agent holds the payload in memory and streams it back.
Every host signs a no-logging agreement
No prompt or output is stored, inspected or reused by the host.
Your idle GPU can earn while you sleep.
Run an OpenAI-compatible local server on Ollama, llama.cpp or vLLM. The agent claims region-local jobs and streams them back.
You choose what runs
Opt in per model class. Your machine only ever claims jobs you enabled it for.
You see every job
Model, tokens, duration and earnings for each job your device ran. Never any content.
Pause or leave in one click
Pausing stops new jobs immediately. Leaving revokes the agent token and keeps your unpaid earnings.
You earn 70% of the job price
Every completed job credits your share. Request a payout once unpaid earnings pass $10.
Two plans. The fee becomes tokens.
Every paid invoice credits the full plan fee to your token balance. It does not expire while you are subscribed. Top-ups of $50, $200 or $1,000 add more.
Builder
For one app moving its open-model calls over.
—/month
- $99 of token balance credited every month, kept while subscribed
- 3 production keys in one region (us or eu)
- Every open-weight model in the catalog at a tenth of list price or less
- Receipt on every call, automatic refund when late
- 4 calls in flight at once
- 30 days of receipts
- Top-ups of $50, $200 or $1,000
Team
PopularFor several apps or environments with separate budgets.
—/month
- $499 of token balance credited every month, kept while subscribed
- 25 production keys across us and eu
- Monthly spend cap per key
- Priority routing: your jobs are picked up first in each region
- 32 calls in flight at once
- 13 months of receipts
- Everything in Builder
Unused balance is refunded if Cheap Tokens ever shuts down, and top-ups are refundable on request. Every account also gets a sandbox key with 25 free test calls, with no time limit.
See what a call costs here.
Run one real call with a sandbox key and read its receipt. Then change two lines and cut the bill.
25 free test calls. No credit card.
Get early access
Leave your email and we'll be in touch.