What is Gonka in simple terms

Today, AI computing is controlled by four corporations: OpenAI, Google, Anthropic, and Meta. They set prices, decide who gets access, and can disconnect an entire country from an intellectual resource with a single decision. This is not a hypothesis — it is an architectural reality: if your business depends on a single company's API, you are hostage to their pricing policy, censorship, and technical outages.

The global AI computing market is valued at over $150 billion and is growing by 30%+ annually. The Stargate project involves hundreds of billions of dollars of investment in building giant data centers. Meanwhile, "second-tier" countries cannot buy GPUs directly from NVIDIA — quotas are limited, and prices are monopolized. Entire regions are left without access to AI simply because they did not make it onto the chipmaker's list of priority customers.

Gonka is creating an open alternative: a network where computing belongs not to a corporation, but to all participants. Any GPU owner can connect to the network and earn. Any user can send an AI request without intermediaries. Price is determined by the market — by the competition of independent hosts, not by a board of directors.

Example: in 2023, OpenAI blocked API access for developers from several countries without warning. Thousands of products built on top of the ChatGPT API were rendered inoperable overnight. In a centralized model, the user has no leverage — you are fully dependent on the decisions of one company. Gonka eliminates this dependency: in a network of 584 GPUs from independent participants, there is no single center of decision-making, no "off" button.

Geopolitical context: in 2024—2025, the U.S. imposed restrictions on the export of advanced GPUs (H100, H200) to a number of countries, including China, Russia, and Iran. This means that entire regions of the world physically cannot access hardware for AI computing. A decentralized network bypasses this problem: GPUs are already distributed around the world, and no government can "turn off" the entire network. Bhutan became the third state in the world to engage in crypto mining at the state level. Uzbekistan is considering connecting state data centers to Gonka. When states begin integration, it is a signal that the technology has moved from the "experiment" category to the "infrastructure" category.

The Gonka team consists of over 50 engineers and researchers working on the protocol, SDK, and network infrastructure from offices in San Francisco and remotely.

The Gonka workflow can be divided into several stages. Owners of powerful GPUs connect their cards to the network by installing specialized software (inferenced CLI) and registering on-chain. Each such machine is called an MLNode — a GPU server ready to execute AI tasks.

When a user sends an AI request, it reaches a Transfer Agent — a specialized intermediary node. The Transfer Agent finds an available MLNode with the required model, routes the request, receives the result, and returns it to the user. All this happens in seconds. The Sprint Consensus (Transformer PoW 2.0) distributes tasks in real-time — each computation simultaneously serves the user and confirms a block on the blockchain. The PoC V2 mechanism verifies that every node operates honestly: 1—10% of tasks are sent for cross-validation by other nodes.

Currently, there are 584 GPUs active in the network (participants in the epoch: 22, MLNodes: 121). Active models include MiniMax M2.7, DeepSeek V4 Flash, and GLM-5.3 Flash, all based on the MoE architecture. The cost of an AI request via Gonka is approximately $0.0069 per million tokens. For comparison, OpenAI GPT-5.5 charges $5 for the same volume of input tokens — which is ~720 times more expensive. The Gonka API is fully compatible with OpenAI — any application using ChatGPT can switch to Gonka in 5 minutes by simply changing the endpoint URL.

Security is ensured by several layers: BLS signatures verify every computation in less than 10 milliseconds, and dishonest nodes lose 20% of their collateral upon detecting discrepancies. All code is open-source on GitHub, and the audit was conducted by CertiK. Unlike centralized providers where security is a management promise, in Gonka, security is an architectural property.

Regarding network models: these are powerful open AI models based on the Mixture-of-Experts architecture — for each request, only a portion of parameters is activated, ensuring high response quality at moderate computational costs. In terms of response quality, they compete with proprietary models like GPT-5.5 and Claude Opus 4.8, but the cost via Gonka is thousands of times lower. The model composition changes via governance votes: in 2026, the network shifted from a single model to several, and Qwen3-235B and Kimi K2.6 eventually gave way to newer ones. The Gonka roadmap includes distributed training of new models via DiLoCo and Confidential Computing for processing sensitive data.

Response speed: a typical request to a network model via Gonka is processed in a few seconds — comparable to ChatGPT. Streaming allows seeing the response as it is generated, without waiting for completion. For developers: the API supports both synchronous requests and Server-Sent Events (SSE) for real-time.

GNK is the Gonka network token, its "fuel." Every AI request is paid for in GNK. Hosts (GPU owners) earn GNK for the work they perform. This is not a speculative meme coin — GNK's value is tied to real demand for AI compute: the more people use the network, the more GNK is needed to pay for requests.

GNK's tokenomics are transparent and fixed in code:

• Total supply: 1 billion GNK — there will never be more.
• 80% (800M) is distributed to hosts for real work — computing AI requests.
• 20% (200M) is reserved for the founders with vesting (gradual unlocking on a schedule).

How distribution works: when a user pays for an AI request, 80% of the payment goes to the host that performed the computation, and 20% goes into the community pool — a fund for ecosystem development, developer bounties, and training new models. Transaction fees (as of October 2026, a regular transfer costs about 0.0002 GNK) are not burned but accumulate in the network's service account.

The token trades on the SafeTrade exchange and via OTC on HEX. The current price is around $0.22. GNK is not yet listed on major CEXs (Binance, Coinbase), but TGE and tier-1 listings are on the project's roadmap.

In practice: if you want to use AI through Gonka, you don't need to buy GNK on an exchange or figure out crypto wallets. There are gateways that accept regular money and convert it to GNK automatically. For developers, the API looks exactly like OpenAI's — the only difference is the URL and the price. And if you want to earn GNK — connect a GPU to the network or invest in a pool.

Verification in Gonka is multi-layered: every inference result goes through Proof of Computation V2 — a cryptographic proof that the GPU actually performed the computation rather than returning random data. Validators compare the hashes of output tensors, and if they diverge, the node loses its rewards for the entire sprint. This makes an attack economically pointless — the cost of faking exceeds the potential reward many times over.

Where to store GNK: the easiest is Karman, a GNK wallet from the joingonka team. It opens like a regular web page, nothing needs to be installed, and the secret phrase and keys stay only on your device. If you plan to delegate GNK to validators, there's GG Wallet (a Chrome extension) or Keplr for that. There's also Gonka.Wallet as a Telegram Mini App and the inferenced CLI wallet for experienced users. For a detailed overview, see the wallet article, and for step-by-step purchasing, see the how to buy GNK guide.

Gonka is open to people with all kinds of experience, budgets, and interests:

• Passive investor: invest from $1 in a mining pool and earn GNK without any technical knowledge. Gonka.Top offers pools from $100 with a 25% referral rate, Ancapex — from $1 with daily payouts (the lowest entry threshold on the market). The operator rents GPUs, sets up nodes, and distributes income — you don't have to configure anything.
• Crypto trader: buy GNK on the SafeTrade exchange or via HEX OTC as an investment in the AI compute market. The current price is around $0.22 with an early-stage market cap. TGE and tier-1 CEX listings are on the horizon.
• Technical user: connect your own GPU server (NVIDIA, from 320 GB VRAM per ML node) and mine GNK directly. This is the most profitable path — the entire reward goes to you with no pool fees. For details, see the mining guide.
• Developer: use Gonka's OpenAI-compatible API for $0.0069/1M tokens instead of $2.50—15 with GPT. Switching takes 5 minutes — just change the endpoint URL. You can also earn GNK through bug bounties and contributions to the open-source code on GitHub.
• Just curious: try the free AI chat on our website, running through the real Gonka network with MiniMax M2.7, DeepSeek V4 Flash, and GLM-5.3 Flash models. No registration, no payment — just see that it works.

Whichever path you choose, we recommend starting with understanding GNK's economics — how the price is formed, what drives demand, and why this token is tied to a real market rather than speculation. For a detailed comparison of all providers and paths, see the "Get GNK" page.

The Gonka ecosystem is growing actively. Beyond the mainnet, several pools are already up and running (Gonka.Top, GonkaPool.ai, Ancapex, CloudMine), along with bare-metal providers (Spheron — users from Russia may have trouble paying through the payment processor) and wallets (Karman, GG Wallet, Gonka.Wallet, Keplr). The community is present on Discord and X (@gonka_ai). The more participants in the ecosystem, the more resilient and valuable the network.