While GPU passthrough remains the standard for high-performance large language models, it often leaves smaller, high-concurrency tasks—such as OCR, speech recognition, and embedding—starved of efficiency. Neutree 1.1 mitigates this by enabling hard-isolated resource partitioning. Administrators can now split GPU memory and compute cycles based on real-time demand, ensuring that diverse workloads operate simultaneously without interfering with one another.

Beyond hardware optimization, the update introduces a unified governance layer for model management. This gateway provides centralized visibility into API quotas, token usage, and security audits, allowing platform teams to enforce access controls across business systems without requiring changes to existing application code. Early adopters, including the global manufacturer Foxconn, are already using the platform to unify their compute and model resource management. The software is currently available as an open-source project on GitHub, with enterprise-grade versions offered directly through Arcfra.