Gonka
Decentralized network for buying and selling AI compute
Last reviewed Jun 25, 2026
Gonka is a decentralized AI compute network. Developers run inference against hosted open-weight models through an OpenAI-compatible API and pay in the network's GNK token, while independent hosts contribute GPU capacity and are rewarded in GNK for serving that inference.
Gonka does not publish public hourly pricing.
Contact the provider directly for a quote.
Pros & Cons
Advantages
- Inference is exposed through an OpenAI-compatible API, so existing client libraries work without rewrites
- Model catalog tracks recent open-weight releases from DeepSeek, Moonshot, and MiniMax
- Consensus design targets useful inference work rather than separate security computation
- Documented onboarding path for GPU owners who want to sell capacity into the network
- Protocol source, white paper, and tokenomics are published openly
Limitations
- Usage is settled in GNK, which requires wallet setup, token acquisition, and exposure to token price movement
- No conventional GPU instance rental — developers buy inference, not machines
- Model availability and capacity depend on independent hosts rather than a single operator
- Hosts must post collateral and operate node infrastructure themselves
Key Features
OpenAI-compatible inference API
Developers send requests through an OpenAI-compatible endpoint and pay for usage in GNK
Compute-as-consensus
A transformer-based proof-of-work mechanism aims to direct nearly all participating hardware at AI inference rather than at separate security computation
Permissionless hosting
GPU owners register as hosts, post collateral, run ML nodes, and are rewarded based on the amount and quality of compute they contribute
Open-weight model catalog
Documented bootstrap procedures for hosting models such as DeepSeek, Kimi, and MiniMax across the network
Self-hosted gateway option
Developers can run their own gateway and broker setup instead of relying on a shared entry point
Cross-chain transfers
Ethereum bridge and IBC routes for moving GNK and USDT in and out of the network
On-chain governance
Protocol changes go through proposals, voting, and delegation by token holders
Pricing Options
| Option | Details |
|---|---|
| Token-metered inference | Inference is paid per request in GNK; per-model rates are published in the network documentation rather than as USD hourly rates |
| Host rewards | GPU hosts earn GNK according to the amount and quality of compute they contribute to the network |
Availability & Support
Support
Documentation site with FAQ and error reference, Discord community, GitHub repository, and a vulnerability reporting process
Getting Started
- 1
Create a Gonka account
Set up a wallet to hold GNK and sign network transactions
- 2
Fund the wallet
Acquire GNK on Uniswap or bridge tokens in from Ethereum
- 3
Connect an OpenAI-compatible client
Follow the developer quickstart to point an existing SDK at the network, or run your own gateway
- 4
Or join as a host
Review the hardware specifications, post collateral, and follow the node setup guide to serve inference
Frequently Asked Questions
Which models does Gonka serve?
Check the pricing table above for the models Gonka currently serves and their per-token rates.
How do I get started with Gonka?
Create a Gonka account, Fund the wallet, Connect an OpenAI-compatible client, Or join as a host
What are Gonka's main advantages?
Gonka's main advantages include: Inference is exposed through an OpenAI-compatible API, so existing client libraries work without rewrites, Model catalog tracks recent open-weight releases from DeepSeek, Moonshot, and MiniMax, Consensus design targets useful inference work rather than separate security computation, Documented onboarding path for GPU owners who want to sell capacity into the network, Protocol source, white paper, and tokenomics are published openly.
What are Gonka's limitations?
Gonka's main limitations include: Usage is settled in GNK, which requires wallet setup, token acquisition, and exposure to token price movement, No conventional GPU instance rental — developers buy inference, not machines, Model availability and capacity depend on independent hosts rather than a single operator, Hosts must post collateral and operate node infrastructure themselves.