- ModelSphere simplifies enterprise large-model deployment and operations and boosts peak throughput by up to 285% (nearly 4x the baseline).
- Building a more open, chip-agnostic foundation for AI at scale.
- Free and open source for enterprise users and available on Github now.
BEIJING, CHINA – Media OutReach Newswire – 16 October 2026 – Phancy Group Co., Ltd. (“Phancy” or “the Group”, stock code: 6682.HK), a leading full-stack enterprise-AI cloud services platform, together with RiseUnion Technology Co., Ltd. (“RiseUnion”), today announced the official launch of ModelSphere, a one-stop, production-grade AI deployment platform.
Free and open source for enterprise users, ModelSphere brings computing resource management, model deployment and runtime optimization together in a single platform. It lowers the barriers to deployment and operations, helping enterprises turn computing power more efficiently into secure, stable AI services ready for real-world business use.
As large models move into commercial deployment at scale, enterprises commonly face low resource utilization, a fragmented chip ecosystem, high deployment barriers and difficult tuning in production environments. ModelSphere aims to consolidate computing resource coordination, model deployment and performance optimization into a unified deployment and management workflow, reducing the duplicated effort enterprises spend on low-level system adaptation, resource scheduling and day-to-day operations.
Phancy believes enterprises should be able to put large models to work without extensive in-house expertise in complex areas such as driver compilation and hardware adaptation. ModelSphere is designed to bridge this gap, enabling enterprises to integrate a wide range of open-source large models into core business processes while allowing technical teams to focus on application development and business needs.
Four Core Advantages of ModelSphere
- Simplified deployment and stable model services: ModelSphere provides a graphical console and pre-built deployment capabilities that reduce the environment-configuration code developers need to write, helping them quickly stand up inference services for large models and simplifying deployment across different chip clusters. Built-in enterprise-grade capabilities, including automatic elastic scaling, prefill-decode (PD) disaggregated deployment, three-tier KV caching, high-availability load balancing and automatic stream resumption after service interruptions, keep model services running stably in production.
- Continuous automated tuning for higher efficiency: The platform’s built-in AutoTune intelligent model scheduling strategy, combined with adaptive load-balancing scheduling that leverages hardware acceleration, improves request-handling capacity and time to first token (TTFT) while lowering inference cost per token. Built-in AutoTune boosts peak throughput by around 285% compared with the baseline, increasing overall performance to nearly four times the previous level and making full use of the available computing power.
- Unified management of diverse chips without repeated adaptation: Integrated with the EngineX inference engine, ModelSphere supports and is deeply optimized for a wide range of AI accelerators. It enables unified and mixed scheduling of all computing resources, including Huawei Ascend, Iluvatar CoreX, Moore Threads and Hygon DCU, so users do not need to rewrite low-level code for each chip.
- Continuous optimization for mainstream models and faster deployment: The ModelSphere team closely tracks iterations in the open-source community and has already adapted and optimized the platform for leading models, including DeepSeek-V4/V4-Flash, Qwen, Kimi, MiniMax and GLM. New models can be deployed quickly after release and tuned to production-grade performance, helping enterprises put the latest models to work in real business scenarios.
How ModelSphere Fits into the AI Stack
At the technical level, ModelSphere works alongside inference engines such as vLLM and SGLang. While inference engines improve computing efficiency on individual servers, ModelSphere handles management and optimization at the cluster-architecture level, including load- and cache-aware request routing, a unified KV cache pool, quality-of-service (QoS)-based automatic scaling, automatic stream resumption after service interruptions, automated performance tuning based on live production data, runtime monitoring and service management. Together, these ensure high performance and high availability in production environments.
By going open source, ModelSphere makes Phancy’s engineering capabilities for large-model deployment and optimization available to more enterprises and developers, and brings together industry partners to continuously expand support for different chips and models. This will reduce hardware adaptation barriers and lower the threshold for large-model adoption, helping enterprises make the most of their existing computing resources and efficiently unlock the commercial potential of heterogeneous computing.
Availability
ModelSphere is now available on GitHub at https://github.com/modelsphere/modelsphere under the Apache 2.0 license.
Hashtag: #Phancy #ModelSphere #AI
The issuer is solely responsible for the content of this announcement.
About Phancy Group
Phancy Group (6682.HK) is a leading full-stack AI cloud services platform, providing comprehensive solutions for the AI 2.0 era. Our offerings include SageAIOS, HAMi vGPU and ModelHub XC, delivering efficient and scalable AI infrastructure with end-to-end capabilities. We provide a complete solution from heterogeneous compute resource management and optimization to the deployment of intelligent agent models. These solutions empower digital transformation across a wide range of industries, supporting our vision of building a large-scale and efficient “Token Factory.”
Guided by the mission of “AI for Everyone” and positioned as the “Navigator of AI,” Phancy Group is committed to becoming a global leader in Artificial General Intelligence.
