Your $5.00 buys 2× as much token usage.
One OpenAI-compatible endpoint for frontier models — DeepSeek, Qwen, GLM and Muse. Prepaid credit, no subscription, no minimum. The same SDK you already use, pointed at a new base URL.
curl https://poolmill.com/v1/chat/completions \
-H "Authorization: Bearer $POOLMILL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek/deepseek-v4.1-flash",
"messages": [
{"role": "user", "content": "Explain CRDTs in two sentences."}
]
}'from openai import OpenAI
client = OpenAI(base_url="https://poolmill.com/v1", api_key="sk-...")
r = client.chat.completions.create(
model="deepseek/deepseek-v4.1-flash",
messages=[{"role": "user", "content": "Say hello."}],
)
print(r.choices[0].message.content)import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://poolmill.com/v1",
apiKey: process.env.POOLMILL_API_KEY,
});
const r = await client.chat.completions.create({
model: "deepseek/deepseek-v4.1-flash",
messages: [{ role: "user", content: "Say hello." }],
});
console.log(r.choices[0].message.content);Half the list price, on the same model IDs.
List price is what the same model costs at published rates elsewhere; effective cost is ours divided by the 2× credit multiplier. Every "you save" column is list ÷ effective — recompute it yourself.
| Model | list price / 1M | PoolMill effective / 1M | you save | ||
|---|---|---|---|---|---|
| in | out | in | out | ||
| DeepSeek v4.1 Flash | $0.3 | $1.2 | $0.075 | $0.3 | 4× |
| Meta Muse Spark 1.2 Contributor | $0.1 | $0.2 | $0.05 | $0.1 | 2× |
| Meta Muse Spark 1.3 Contributor | $0.1 | $0.2 | $0.05 | $0.1 | 2× |
List prices: published rates for identical model IDs, captured 2026-09-21. Rates move — the PoolMill column is always live on the pricing page.
Fifty times cheaper on the tokens that repeat.
Most of the cost in a real application is the same system prompt and context, sent again and again. We bill those tokens at the cached-input rate instead of the input rate — and we show you the split on every request.
{
"usage": {
"prompt_tokens": 1233,
"completion_tokens": 5,
"total_tokens": 1238,
"prompt_tokens_details": { "cached_tokens": 1024 }
}
}A real response. 83% of the prompt was served from cache, so it was billed at the cached rate.
You buy credit once. It does not expire on a cycle.
There is no seat, no platform fee and no minimum spend. The multiplier is what you get in metered usage for what you pay, priced at the providers' own published rates — not a marked-up copy of them.
What a key can reach
A key is scoped to the models in the package you bought. Anything else is
rejected with 403.
An endpoint should be the boring part of your stack.
No sales call, no waitlist, no approval step. Buy credit, create a key, change one URL — and get back to the thing you were actually building.
Your existing SDK, pointed somewhere else.
The OpenAI SDK works unchanged — streaming, tool calling and JSON mode behave as
they do upstream. The Anthropic SDK works too: the same host serves
/v1/messages.
from openai import OpenAI
client = OpenAI(base_url="https://poolmill.com/v1", api_key="sk-...")
r = client.chat.completions.create(
model="deepseek/deepseek-v4.1-flash",
messages=[{"role": "user", "content": "Say hello."}],
)
print(r.choices[0].message.content)Two minutes to first token.
Get credit
Pick a package on the pricing page. Your balance is available immediately — no approval step.
Create a key
Sign in to the console and create an API key. It is scoped automatically to the models in your package.
Point your SDK at us
Change the base URL to https://poolmill.com/v1 and use your
key. That is the whole migration.
The ones that come first.
How is this cheaper than going direct?
We buy capacity in advance at a discount and pass part of it on as a credit multiplier. You pay $5.00 and receive 2× of usage at the providers' own published rates — the rates on our pricing page are the upstream list rates, not a marked-up copy.
Do I need to change my code?
Only the base URL and the model name. The API is OpenAI-compatible, so existing SDKs, streaming and tool-calling code work as they are. See the quickstart.
What happens when my credit runs out?
Requests return HTTP 429 with a budget message.
Nothing is deleted and no card is charged — add credit and the key works again
immediately.
Is prompt caching really that much cheaper?
Yes, and it applies automatically. Cached input on DeepSeek v4.1 Flash is $0.003 per 1M against $0.15 for standard input — 50× cheaper. Long, stable system prompts and agent loops benefit the most.
Do you store my prompts?
No. We record token counts and cost per request so your usage can be metered and audited; the request and response bodies are not persisted in our records.