Performances
RD Popularity Index
RD Growth Index
Comparison
Fundraising
Follow Updates
Follow Lists
People
About Gonka
Gonka is a decentralized GPU network that connects global hardware providers with AI developers. It offers an OpenAI-compatible API for running open-source models, with payments settled in its native token GNK. The network is built on a Proof-of-Work model, which the project argues is better suited for AI compute than the Proof-of-Stake approach used by most crypto networks.
Its core value proposition is straightforward: developers can access AI inference at significantly lower cost than centralized cloud services, while GPU owners can monetize idle hardware. The project emphasizes that it does not retain control over the network, with governance delegated to the community.
The pain point it addresses is the high cost and centralized control of AI compute. Many developers who rely on centralized inference APIs face escalating bills as their usage grows. Gonka positions itself as an open alternative, allowing anyone to contribute or consume compute without permission.
In the past six months (March to September 2026), Gonka has faced significant challenges. The network experienced dozens of DDoS attacks, with the most intense period occurring between late December 2025 and January 2026. During peak attacks, several nodes went offline daily. More critically, the initial incentive mechanism allowed attackers to profit by taking down other miners—if 30% of miners were knocked offline, the attacker's rewards would increase by 30%. This flaw was fixed through a community vote, removing the direct economic incentive for attacks. The broader crypto market downturn also reduced GPU participation, though the network still maintains approximately $200 million in active hardware assets. The project has also been in discussions with government officials about using decentralized compute for national AI infrastructure, though these remain exploratory.
Gonka launched in August 2025 as a decentralized AI compute network, using almost 100% of resources for AI workloads (unlike PoS), with a 180-epoch zero-cost inference period and GNK rewards.
In early 2026, governance approved v0.2.9, enabling PoC v2 as the sole weight mechanism, unifying model validation (Qwen3-235B) and entering single-model operation. Network hashpower reached ~14,000 H100-equivalent by Feb 2026.
After repeated DDoS attacks, the community voted to change incentive design so attacking other miners no longer yields direct economic gain, significantly reducing attack motivation.
Proposal 78 made MiniMax-M2.7 the sole PoC model; later DeepSeek V4 Flash became an active model group (epoch 360) with calibrated weight conversion coefficient 0.214.
Governance passed v0.2.15, introducing devshard v4 runtime for high availability, and later began gradual settlement to raise security — a key step toward mature decentralized infrastructure.
Over the past six months, Gonka has advanced its decentralized compute network through two major releases. v0.2.12 introduced multi-model PoC support, enabling parallel inference for models like Kimi K2.6, along with devshard runtime isolation, randomized MLNode selection to close hardware-swap exploits, and resolution of all Certik audit findings. v0.2.15 completed the removal of legacy inference APIs, added an independently scalable edge-api service for read-only traffic, and enabled high-availability devshard v4 deployments with multiple replicas behind a router on shared Postgres.
Looking ahead, the official roadmap outlines three phases: first, becoming the leading decentralized AI compute protocol; second, competing directly with neo-clouds like Nebius and CoreWeave; and third, evolving into a frontier-scale, decentralized compute layer for global AI infrastructure. The community-approved plan emphasizes verifiable inference, permissionless access, and governance-weighted Proof of Compute, with a long-term goal of onboarding tens of millions of H100-equivalent GPUs into the network.
Co-founders: Daniil Liberman (CEO), David Liberman (CTO), Anna Liberman, Masha Liberman. All are siblings from the Liberman family. Daniil and David founded Kernel AR (acquired by Snap Inc. in 2016), then Product Science (raised $18M). Anna and Masha joined later. Masha served as Sr Manager at Snap's Mobile Performance Lab and Head of Content for 3D Bitmoji. The team previously built Sibilant Interactive (game studio) and Concept Space (animation studio). Gonka.ai was incubated by Product Science research team.