From $2,000/mo OpenAI-compatible
Run the top open-source model at the best possible price
Join a shared pool on top-tier GPU servers and get compute below market price
How it works
Pick the pool
Pick the pool with the model you want
One pool, one model, one server. Its throughput and context window are published
Commit a budget
Commit a budget, from $2,000 a month
The app shows the share of the server and the tokens per month that buys
Wait until the pool is filled
Funds sit in escrow until launch
On-chain escrow, refundable at any time before the server starts
Join the pool
The server runs within 24 hours from the time the pool is closed
Monthly renewal, or a single month
How you get
the cheapest tokens
A share of a real server
You get a part of a real server and all the compute it gives
Compute becomes
AI tokensYou get all tokens that your part of the server produced
Unspent tokens
are resoldEvery token you do not use, we resell on dex.do
The proceeds come back to you
We resell what you don't use and it comes off your next month
Nobody sees your prompts, answers or spend
on it
Anthropic and OpenAI train on consumer-tier prompts, and keeps that data for five years
In September 2026 OpenAI could not rule out that de-identified data from two mathematicians' use of its products had improved its models. Anthropic moved Free, Pro and Max to opt-out training in September 2025, with five-year retention
Nobody
Open weights on your share of the server — no vendor, no training clause
demand it
Every operator under PRC jurisdiction must assist state security organs — DeepSeek, Qwen and Moonshot included
Cybersecurity Law art. 28 binds network operators; CAC Interim Measures art. 19 binds generative-AI providers specifically
Nobody
No provider in the loop, no foreign jurisdiction over your prompts
your spend
Your provider knows exactly how much you buy and when
Shielded payments
Amounts and counterparties stay hidden
Everything you need to know about gosh.ai DePool
How is the price set?
You commit a monthly budget and we run the models on our own tuned hardware, which lands the per-token price below market. Fixed for the term, quoted per deal
What do you mean by shared pools?
Your budget joins a shared pool running on top-tier GPU servers. Aggregated demand is what brings compute below market, and you get a fixed per-token price for the term
Do I change my code?
No. It is OpenAI-compatible. Point your base URL at us
What is the minimum?
$2,000 a month. Your budget sits in on-chain escrow and is released as you use it