Gonka v0.2.9 mainnet upgrade completed, PoC v2 officially launched

According to Odaily, the decentralized AI computing power network Gonka has completed its v0.2.9 mainnet upgrade. This upgrade was approved through on-chain governance voting and was officially executed at block height 2,451,000. The network has now fully switched to PoC v2 as the weight allocation mechanism, with the original PoC logic being simultaneously phased out. This upgrade marks Gonka's entry into a more mature stage in terms of computing power verification mechanisms and network governance.

After the upgrade takes effect, Confirmation PoC becomes the authoritative source of network results, further enhancing the verifiability and certainty of computing power contributions. At the same time, the network has entered a single-model operation phase, reducing heterogeneous computing noise by adopting a unified model and verification standard, thus providing a more stable infrastructure environment for decentralized AI inference and training. Currently, only ML Nodes running Qwen/Qwen3-235B-A22B-Instruct-2507-FP8 and using PoC v2 compatible images can participate in weight calculation. The transition period from Epoch 158 to 159 will be the first complete operational phase after PoC v2 is enabled.

According to real-time data from GonkaScan, as of February 2, 2026, Gonka's total network computing power is close to 14,000 H100 equivalents, already exhibiting the scale characteristics of a national-level AI computing cluster. Compared to the approximately 6,000 H100 equivalents at the time of Bitfury's $50 million investment announcement in early December 2025, the network's computing power has shown a monthly growth rate of about 52%, leading the pace among similar decentralized computing power networks.

In terms of computing power structure, high-end GPUs such as NVIDIA H100, H200, and A100 account for more than 80% of the network's total computing power, demonstrating Gonka's significant advantages in high-performance computing resource aggregation and scheduling. Currently, network nodes cover about 20 countries and regions across Europe, Asia, the Middle East, and North America, laying the foundation for building a global, single-point-of-failure-resistant AI computing infrastructure.