Sprint: How Gonka's Consensus Works

Consensus is the mechanism by which all nodes in a network agree on the current state of the blockchain. Without it, every node would see its own version of the "truth," and the network couldn't function. Blockchain history has seen three generations of consensus:

• PoW (Bitcoin, 2009) — miners grind through meaningless SHA-256 hashes. Secure, but wasteful: 100% of the energy goes into a "digital lottery."
• PoS (Ethereum, 2022) — validators lock up tokens as collateral. Energy-efficient, but it sacrifices decentralization: large stakers (Lido, Coinbase) control a significant share of the network.
• Sprint / PoUW (Gonka, 2025) — a third path. GPUs perform real AI computations that simultaneously serve users and confirm blocks.

Sprint is a Transformer PoW 2.0, Gonka's unique consensus. The name captures the essence: the network's work is organized into sprints (epochs), each of which is a cycle of execution, verification, and reward distribution. Unlike Bitcoin, where "work" means iterating through random numbers, in Sprint work means a forward pass through one of the network's models (for example, MiniMax M2.7).

Network scale: Gonka currently runs 1,053 GPUs (participants in the epoch: 28, ML nodes: 246). Each ML node is a GPU server with at least 320 GB of VRAM, capable of handling neural network requests. The network's heaviest MoE models need more — hundreds of gigabytes per replica. Sprint coordinates all these resources in real time, distributing tasks and tracking each participant's contribution.

In the Gonka network, participants perform different roles, and Sprint coordinates their interaction. Hosts (miners) provide GPUs for inference and model training — they earn GNK both from the genesis emission and from payment for completed requests. Transfer Agents are specialized dispatcher nodes that accept incoming AI requests, verify cryptographic signatures, and route requests to suitable ML nodes based on load, available model, and latency. Validators provide cryptographic auditing — they verify that nodes performed computations honestly. All roles are economically incentivized through GNK rewards, and Sprint coordinates them without central control.

The network is fully permissionless — any owner of a suitable GPU can connect without KYC by installing cosmovisor and MLNode. Task routing between nodes is based on hardware specifications, availability, and node reputation. Voting power in the network (Proof of Compute) is determined by the volume of computational work: "one unit of compute = one vote." This is fundamentally different from PoS, where voting power is determined by capital.

The key difference from its predecessors: Sprint is not an "empty" protocol tied to abstract difficulty. Every block contains proof of work actually performed — neural network responses that were sent to users. This creates a direct link between network security and its usefulness: the more AI requests the network processes, the more secure it becomes.

To understand why Sprint is an evolution rather than just 'another blockchain', it is useful to compare it with classic Bitcoin Proof of Work:

The main difference lies in the source of value. Bitcoin's value relies on 'digital gold' — limited supply and consensus on value. GNK's value is tied to the real demand for AI computing. When someone sends a query via Gonka API, they pay GNK for a specific result — a neural network response. This is utilitarian demand, not speculative.

The second critical difference is scalability. Bitcoin cannot scale: more ASICs = more wasted energy, but not more transactions. In Sprint, every new GPU increases network throughput — more queries are processed in parallel, and blocks are formed faster. Network growth does not lead to energy waste, but increases its useful power.

The third difference is hardware versatility. Bitcoin ASIC miners are useless outside of mining — they are specialized chips that only know how to calculate SHA-256. The GPUs used in Gonka (H100, H200, A100) are universal computing accelerators. If a host decides to leave the network, their hardware retains full value — it can be used for rendering, scientific calculations, model training, or sold on the secondary market.

The fourth difference is the reward economic model. In Bitcoin, the reward is fixed and halves every ~4 years (halving), regardless of how much real work is performed. In Sprint, rewards are distributed proportionally to the Active Proof-of-Compute weight — the more AI queries a host processes, the more GNK they receive. Additionally, hosts receive two parallel income streams: tokens from the genesis supply (80% of the total 1 billion GNK is distributed to hosts) and payment from users for inference (80% of each AI query). To maximize rewards, hosts lock GNK as collateral — without collateral, the node's weight is reduced by 5 times. This creates an economic incentive for long-term participation in the network, rather than 'pump and dump'.

Sprint security has been confirmed by an audit from CertiK — a leading Web3 audit firm, which completed its review in September 2025. The project has raised ~$80M in investment from Coatue, Bitfury ($50M Series B), Insight Partners, and Benchmark — this is institutional confirmation of the viability of the Sprint model as a next-generation consensus.

Let's walk through the full lifecycle of an AI request on the Gonka network — from the moment a user hits "Submit" to receiving the response and distributing rewards:

1. Request: the user (or an application) sends a standard POST /v1/chat/completions through an OpenAI-compatible API. The request is signed with the wallet's cryptographic key — this proves the sender has the funds to pay.
2. Routing: the request reaches a Transfer Agent — a specialized dispatcher node. The Transfer Agent verifies the signature, identifies the required model, and finds an available ML node with the right specs (enough VRAM, the right model loaded, minimal latency). Multiple Transfer Agents run simultaneously for fault tolerance.
3. Inference (compute phase): the selected ML node performs a forward pass through the chosen network model — for example, MiniMax M2.7. The GPU streams the response token by token. The user receives the answer in real time — latency is minimal.
4. Verification (PoC V2): while the request is being processed, the network verifies node honesty. 1—10% of tasks are randomly sent for re-execution by another node. The results are compared. If they match — both nodes confirm their honesty. If not — arbitration begins, and the dishonest node loses 20% of its stake. BLS signatures make it possible to verify results in under 10 milliseconds.
5. Block formation: at the end of an epoch, all proofs of completed work are aggregated into a block. The block contains: hashes of completed requests, BLS verification signatures, and contribution data for each node.
6. Reward distribution: GNK rewards are credited in proportion to each node's contribution. 80% of the inference fee goes to the host that processed the request. 20% goes to the Community Pool — a fund for ecosystem development (model training, grants). Hosts additionally receive tokens from the genesis emission — these are two parallel income streams.

The whole process takes seconds. The user never sees the blockchain mechanics — for them it's just a regular API call to a neural network, the same as calling ChatGPT, only many times cheaper. The current network inference price is around $0.005 per million tokens, versus $2.50—15 per million tokens at OpenAI.

Dynamic pricing is another Sprint feature. The inference price is recalculated every block based on network load. In the stability zone (40—60% utilization) the price stays the same. Below 40% — the price drops automatically to attract users. Above 60% — it rises, incentivizing new GPU connections. The maximum change is 2% per block. This creates a market mechanism where supply and demand balance automatically, with no manual intervention.

Security through DiLoCo: beyond inference, the network's roadmap includes distributed model training — GPU clusters around the world train locally and synchronize once every ~1000 steps via the DiLoCo protocol. This approach makes it possible to train AI models without gathering all GPUs into a single data center; for now this is a research direction. Multi-model inference is already live: hosts serve different models depending on their GPUs. Sprint turns Gonka from an inference network into a full-fledged AI platform.

Protocol-level security: to protect against consensus attacks, Gonka runs Guardian nodes — 3 special nodes that control 34% of the votes. This guarantees that an attacker cannot seize control of the network, even with significant computational resources. Guardian nodes were introduced in v0.2.7 (January 2026) as an extra security layer during the network's early stage. As the number of independent nodes grows, the role of Guardian nodes will gradually diminish through on-chain governance — a management process in which voting power comes from a host's PoC weight (since v0.2.16 — no higher than what was confirmed in the previous epoch), not from token balance. This is standard practice for young blockchain networks: start with a controlled launch and gradually move toward full decentralization.