Gonka 0.2.12, which rivals both Kimi and DeepSeek, is ending the "wild west" era of decentralized computing power.

In April 2026, tech giants like Microsoft, Google, and Meta are expected to invest over $600 billion in AI, pushing centralized computing to its limits. DeepSeek V4's launch significantly reduces inference costs, ushering in the 'Inference Economy.' Against this backdrop, decentralized AI network Gonka releases version 0.2.12, enforcing model cleanup, deeply refactoring vLLM to support frontier models like Kimi and DeepSeek, and introducing multi-model proof-of-computation, voting delegation, and unified multipliers to ensure fair verification. Additionally, Blackwell architecture optimization boosts B200 GPU earnings above H100. Gonka aims to establish industrial-grade order in decentralized inference, driving DeAI toward mass adoption.

"The delivery gap in AI infrastructure is structural, not temporary." In late April 2026, as Microsoft, Google, and Meta released their Q1 financial reports , Wall Street's focus shifted from "AI visions" to "AI bills." According to the latest statistics from Reuters on April 28, the four tech giants' AI investment this year will exceed $600 billion . This unprecedented expenditure is squeezing cash flow and testing the market's patience.

Just a few days ago, the release of the DeepSeek V4 preview version once again ignited the industry: this model, known for its "extreme cost-effectiveness," not only pushed inference prices to rock bottom but also heralded the full arrival of the "inference economy" in the AI 2.0 era. Under the dual pressure of extreme hunger for computing power and extreme sensitivity to inference costs, decentralized AI (DeAI) is ushering in its "industrialization" singularity.

Recently, Gonka Protocol released a landmark version 0.2.12 update . This is not only a technical iteration, but also a "clean-up operation" targeting the chaos in DeAI. Through the reconstruction of the Multi-Model Proof-of-Capacity (PoC) protocol, Gonka, while achieving "full compatibility" with top models such as Kimi and DeepSeek, is ending the old era of disorder with industrial-grade standards.

I. April Observations: The $600 Billion "Computing Power Siege" and DeepSeek's Price War

In April 2026, two major events occurred in the AI infrastructure field, enough to reshape the landscape:

1. The arms race among tech giants has entered a deeper phase.

Financial reports released at the end of April showed that Microsoft's annual capital expenditures had been revised upward from $161.6 billion to $180 billion. S&P Global predicts that global cloud vendors' combined Capex will surge from $379 billion to $622 billion by 2026. This scale of investment reflects that centralized computing power supply is approaching its physical limits.

2. DeepSeek V4's "Inference Equality"

On April 24th, the release of DeepSeek V4 completely changed the game. It not only rivals top-tier models in performance but also reduces inference costs several times over through extreme engineering optimizations. This means that future competition will no longer be about "who has the bigger model," but rather "whose inference is cheaper and more stable." In this context, DeAI networks capable of providing low-cost, highly flexible computing power have become an industry necessity.

II. Clearing the Field of Efficiency: From "Extravagant Shows" to "Industrial-Grade Standardization"

Prior to version 0.2.12, the DeAI network was often criticized as a "laboratory toy." Gonka completely changed this situation through two core changes:

1. Forced model cleaning: The first step in eliminating false models and retaining the true ones.

According to the 0.2.12 upgrade announcement, Gonka adopted a strong "model cleaning" strategy: all models that have not been approved by governance will be forcibly taken offline. The network will focus on benchmark models such as Qwen and the newly approved Kimi 2.6. This seemingly scale-down action is actually to ensure that every unit of computing power in the network is real, verifiable, and conforms to industry standards.

2. Deep Restructuring of vLLM: Breaking Through the "Technical Ceiling" of Cutting-Edge Models

To support cutting-edge models with different architectures such as Kimi 2.6 and DeepSeek V4, Gonka has completely rebuilt its vLLM framework. Through architectural optimization, nodes can complete verification while keeping the model resident in GPU memory, completely solving the pain point of slow loading of large models and enabling decentralized inference to achieve commercial-grade response speeds.

III. The Cornerstone of Trust: Multi-Model PoC and "Delegation of Voting Rights"

In a permissionless environment, trust must be built on extremely high manipulation costs.

1. Solving the "verification trilemma"

When faced with massive models like Qwen-235B or Kimi, which are hundreds of gigabytes in size, nodes cannot switch to perform verification in real time. Gonka 0.2.12 introduces an innovative "voting delegation" mechanism: during the Proof-of-Computation (PoC) process, nodes use their own voting rights or delegated voting rights to cross-validate participants running the same or different models.

2. Unified Multiplier: Fairly measures every unit of computing power.

Because different models generate different computational units on the same hardware, version 0.2.12 introduced a "uniform multiplier" mechanism. This ensures that regardless of whether a miner is running Qwen or Kimi, their actual hardware contribution is fairly reflected, thus protecting the fairness of the entire ecosystem during the token subsidy phase.

IV. Hardware Justice: Redefining the Profit Ceiling for B200

Technology must not only address trust issues, but also correct distribution failures.

In previous versions, due to a lack of targeted optimization, the performance of expensive NVIDIA Blackwell (B200) graphics cards was not significantly different from that of the H100. Gonka 0.2.12, through specific optimizations for the Blackwell architecture, unleashes the true performance of the B200, enabling it to surpass the H100 in profitability once again. This pursuit of "hardware justice" will attract the world's top computing resources to Gonka.

Conclusion: A "Declaration of Order" for the DeAI Era

Just as early Bitcoin ended the chaos of electronic cash through the PoW mechanism, Gonka 0.2.12 is establishing the same order in the realm of decentralized reasoning.

From Big Tech's $600 billion bill at the end of April to the inference price war initiated by DeepSeek V4, the second half of the computing power war has begun. Gonka is telling the market: the era of grassroots heroes is over. In the marathon of AI 2.0, only "order" built on core technology and rigorous logic is the only path for DeAI to achieve large-scale commercialization.

About Gonka

Gonka is currently the decentralized AI network with the largest number of GPUs, providing developers and researchers with permissionless access to computing resources while rewarding all participants through its native token, GNK. The project successfully raised $18 million in 2023 and an additional $51 million in 2025. Investors include OpenAI investor Coatue Management, Solana investor Slow Ventures, Bitfury, K5, and partners at Insight and Benchmark. Early contributors include leading companies in the Web 2-Web 3 fields such as 6blocks, Hard Yaka, and Gcore.