Open-model inference, verifiable
Open-model inference
you can audit.
Point your OpenAI client at Orogen and your existing code runs unchanged, at roughly $0.40 per 1M output tokens for Llama-3.1-70B. Every response comes back with a signed receipt you can verify yourself, and validators re-run a sample of jobs on identical hardware to catch anyone cutting corners.
- OpenAI-compatible HTTP + SSE
- A signed receipt on every response
- Validator replay on identical hardware
Why Orogen
Inference you do not have to take on faith.
Most inference networks rent you a GPU and ask you to trust whatever comes back. You cannot tell which model actually answered, whether it was the quantization you paid for, or whether the hardware was what it claimed, so you reconcile an invoice and hope it was honest. Orogen works the other way around. The network verifies the work and hands you the proof, so you check the result rather than take the operator's word for it.
A receipt you can check yourself
Every response carries a signed receipt that binds the model, your input, the output, and a hardware attestation into one record. You verify it locally in a single call before you trust the result.
The network checks the work for you
Validators independently re-run a sample of jobs on hardware identical to the operator's. Any mismatch between what an operator claimed and what the replay produces opens a challenge.
Paid in dollar terms for real work
Operators are paid per completed, verified job at oracle spot, denominated against the dollar, with most of each payout auto-swapped to a stablecoin. Pay tracks served demand, not token price.
Issuance tracks revenue
Customers pay in stablecoins, the gateway burns OROG to settle each job, and new OROG is minted only against work that validators have verified. Supply follows real inference revenue.
How it works
A loop between you, the network, and the chain.
Verification is a loop, and you sit at the start of it. Read the full walkthrough in how it works.
- 1
You verify
You send a request, receive an answer and a signed receipt, and verify that receipt locally in a single call.
- 2
The network checks
Validators replay a sample of jobs on identical hardware and flag any mismatch, which opens a challenge.
- 3
The chain records
Each job's commitment is stored on-chain, so a disputed result has a record to settle against rather than two parties arguing from memory.
For builders
Drop-in for the OpenAI code you already wrote.
Set your base URL to the Orogen gateway and your existing code keeps working. There is no new SDK to learn and no rewrite to schedule. Reach for the Orogen SDK when you want to pick a hardware tier per request or pull the receipt object directly.
from openai import OpenAI
client = OpenAI(
api_key="orog_...",
base_url="https://gateway.orogen.network/v1",
)
response = client.chat.completions.create(
model="llama-3.1-70b-instruct",
messages=[{"role": "user", "content": "Summarise this contract clause..."}],
)Pricing
Roughly $0.40 per 1M output tokens.
That is the standard datacenter tier for Llama-3.1-70B, against $10 to $15 for comparable closed-model APIs. You pick the hardware tier per request, you pay only for the tokens you use, and there is no subscription floor.
For operators
Run an open-source inference server with three additions: hardware attestation, deterministic kernels, and the receipt sidecar. You are paid per completed job at oracle spot, denominated against the dollar, with most of each payout auto-swapped to a stablecoin. Pay tracks verified work, so honest operators are not diluted by emissions handed out for nothing.
Why the token holds up
Customers pay in stablecoins, and the gateway burns OROG to settle each job. New issuance is minted only against work that validators have verified, so issuance tracks real inference revenue rather than a fixed inflation schedule. There is no foundation mint discretion and no halving.
Sourced from prior networks
What we learned from the networks before us.
Orogen's design comes from studying what prior networks got right and what they left on the table.
Bittensor SN3 showed that a global operator base already exists, with more than 70 peers across three continents running on commodity links.
Phala showed that TEE attestation works at scale, routing more than a billion tokens a day for OpenRouter clients.
Akash showed that burn-and-mint can move issuance toward real revenue instead of a fixed inflation schedule.
Orogen anchors all three to verified physical work. Like the orogeny that slowly builds a mountain range under pressure, durable infrastructure is the product of accumulated verified work rather than announcements.
Build on it, or get paid to run it.
Start building against the live gateway with the OpenAI client you already have, or run a node and get paid in dollar terms for verified work.