Short answers to the questions we get most about Gonka and about reading it on GNKScan. Every answer links to a longer write-up if you want the full mechanics.

What is Gonka?

Gonka is a blockchain for decentralized AI compute. Hardware providers earn consensus power by proving how much ML compute they control, then spend most of each epoch serving paid AI inference. The chain is a modified Cosmos SDK application whose validator power comes from computational results rather than staked tokens. The long version is in What Is Gonka?

How does Gonka reach consensus without staking?

Through proof of compute. Each epoch opens with a structured competition — the whitepaper calls it a Sprint — in which every participant's GPUs generate transformer-based proofs seeded by a recent block. Other participants cross-validate a sample of those proofs, and accepted results become consensus voting power via a function called SetComputeValidators. No tokens are bonded for consensus at all. Details in How Gonka's Proof of Compute Works.

What is an epoch on Gonka?

An epoch is the network's unit of time: a span of blocks with a fixed internal schedule of phases — proof generation, validation, a long stretch of inference service, then settlement at the boundary. Validator rotation, payments, reward vesting and model assignments all happen at epoch boundaries. See Gonka Epochs Explained.

What is GNK?

GNK is the network's native token. Participants earn it two ways at each epoch's settlement: work coins (escrowed user fees for inference they served) and reward coins (newly minted GNK split by each participant's share of the epoch's work). The full economic design is in Gonka Tokenomics.

Does Gonka charge gas fees?

No. Ordinary transactions are free — the chain's configuration sets transaction fees to zero. Transactions still carry gas accounting (used and wanted), which you can see on any block page, but nobody pays it. All economic weight rests on inference fees instead.

How is AI inference priced on Gonka?

Fee = (prompt tokens + actual completion tokens) × units-of-compute-per-token × unit-of-compute price. The per-token figure is a per-model constant set at registration; the unit price is a weighted median of participant proposals, updated each epoch. The maximum cost is escrowed up front and the unused remainder refunded after the response completes.

Which AI models does Gonka serve?

Models are registered through on-chain governance, each with its own units-of-compute-per-token figure. The current on-chain registry lists MiniMax M2.7, Kimi K2.6 and GLM 5.2. The models page tracks what is registered and being served.

How many participants does Gonka have?

Two very different numbers get quoted. As of early July 2026 (around epoch 318), the active set — participants actually doing work in the current epoch — is roughly 35, operating on the order of 196 ML nodes and roughly a thousand live GPUs. The cumulative count of participant addresses registered since genesis is in the thousands. GNKScan's participants page reports the active set, because that is what secures the chain.

What hardware runs the network?

The live GPU fleet is dominated by NVIDIA data-center accelerators — B200, H100, H200, B300 and A100 among them. You can see the compute breakdown on the GNKScan home page and in analytics.

What does a Gonka address look like?

Gonka uses Cosmos-style bech32 addresses with the gonka1 prefix. Paste any address into the search bar on GNKScan to see its transaction history.

Why does a field on GNKScan show "—" instead of a value?

Because the value was not available from a source we trust. GNKScan draws on our own indexer (complete block data) and the network's public indexer (per-block transactions, but no block hash or proposer). When only the public source covers a block, the missing fields render as an em dash rather than a fabricated value. The policy is explained in How to Read a Gonka Block.

Why do explorers disagree about participant or node counts?

Usually one of two mix-ups: active-this-epoch participants versus cumulative registered addresses, or Gonka's two parallel power systems — consensus power in the staking module versus epoch-group power in the inference module. They synchronize at epoch boundaries but can legitimately differ mid-epoch. What Is Gonka? covers the first distinction and How Gonka's Proof of Compute Works the second.

What happens if a participant cheats or goes offline?

Validation is designed to catch it: proofs are cross-checked by sampled re-execution, and inference itself is validated at a rate that scales with volume. Under Tokenomics V2, most of a participant's weight must be backed by GNK collateral, which is slashable — 20% by default for malicious behavior, 10% for failing an epoch's participation requirements. See Gonka Tokenomics.

Where should I start reading the chain?

Start at the current blocks and the live transaction feed, with How to Read a Gonka Block open next to them. Once you can tell a proof-of-compute burst from ordinary inference traffic, the rest of the chain reads itself.