Running a model with around 744 billion parameters is a challenging task for any single graphics card. To address this, the AI platform Virtuals Protocol has teamed up with Leyten to implement an innovative solution. They have integrated Leyten’s shard engine, a system built to distribute the computational workload of large-model inference across multiple GPUs within a network. Their initial goal is to process GLM-5.2, a recently released open-weight model by Z.ai under the MIT license, which became available on June 16, 2026. This collaboration provides Virtuals with a pathway to execute frontier-scale AI inference, eliminating the need for centralized cloud computing resources or vast GPU clusters.
#Why is GLM-5.2 Significant?
GLM-5.2 represents a massive leap in AI model capability. With approximately 744 billion total parameters, it is designed to activate only 39 to 40 billion parameters per token during processing. This mixture-of-experts architecture allows the model to operate efficiently, activating just a fraction of its stored knowledge for any given task. This design significantly mitigates compute costs while managing the model's enormous size. It also features an impressive context window of one million tokens, five times bigger than its predecessor, GLM-5.1. Released to subscribers on June 13, 2026, and later to the public, GLM-5.2 can be used, modified, and commercially deployed by anyone due to its open license.
#How Does Leyten’s Shard Engine Work?
Leyten’s shard engine presents a unique solution for handling large AI models. It employs pipeline-parallel inference, which effectively divides a large model into smaller segments that can be distributed across multiple GPUs connected over a network. This method eliminates the need for any single GPU to manage the entire model in its memory, resulting in more efficient processing and flexibility.
#What Role Does Virtuals Protocol Play in the AI Agent Market?
Virtuals Protocol is positioned within the cryptocurrency sector, specifically targeting the development and monetization of on-chain AI agents. These autonomous digital entities are capable of performing transactions, executing tasks, and interacting with blockchain protocols. The native token of this ecosystem is VIRTUAL.
The introduction of GLM-5.2 aims to offer a competitive alternative to proprietary advanced models while operating at substantially lower costs. Its capabilities, featuring a million-token context and robust coding potential, create a solid foundation for making autonomous AI behaviors a reality rather than a distant goal.