SINGAPORE — Arcfra, the Singapore-based developer of enterprise cloud and AI infrastructure software, has announced Neutree 1.2, a new release of its enterprise AI platform that widens the range of models it can serve and streamlines how companies plan resources and manage operations. The announcement went out through PR Newswire on Wednesday, September 16, 2026, at 11:42 pm GMT-4 — which is September 17 in Singapore. The release is identified as a paid press release.
Neutree 1.2 is available now, according to the company.
A new engine for the models that aren’t chatbots
At the center of the release is Flex Engine, an Arcfra-developed inference engine that supports non-LLM models such as MinerU and PaddleOCR. That matters because a great deal of enterprise AI work never touches a large language model at all. Document parsing, optical character recognition, embedding, rerank and classical machine learning models all do real work inside companies, and they have often lived on separate platforms from the language models.
Neutree 1.2 builds on standard inference engines such as vLLM and SGLang. On top of that foundation, the platform lets users manage language, multimodal, embedding, rerank, document parsing, OCR and machine learning models in one place, then publish and manage service calls through Neutree’s unified model gateway.
Arcfra says this reduces duplicated platform work and accelerates the path from model validation to business use.
Fewer failed deployments, less wasted memory
The company also added automatic KV-cache and GPU memory calculation. The feature parses a model’s structure and works out the KV-cache a deployment will need based on the user’s context window and concurrency settings, then recommends the GPU memory required. The stated aim is to cut the risk of failed deployments while avoiding unnecessary GPU memory waste — a practical concern for any team that has watched a rollout stall on resource planning.
Visual model registry management is new as well. It supports unified management of public and private models, showing source, connection status, visibility, model count, storage usage and update time in a single interface, along with each model’s parameter count, size and precision.
API key management got a rework too. Users can now group multiple API keys by business project and identify ownership through the project name and description. Keys display their workspace, status, usage, rate limit, supported models and creation time, with search and filtering.
Arcfra describes the shift as moving from a simple key list to a business-oriented view that clarifies ownership across teams.
Customers in manufacturing and shipbuilding
Several customers have started using Neutree, according to the announcement. A manufacturing leader used its existing Arcfra infrastructure as a production-grade AI foundation to streamline model delivery and unify compute and model resource management. A shipbuilding customer brought compute management, model management and the model gateway into a single platform, enabling faster model rollout and unified operations.
The press release did not name the customers or provide financial terms. Taken together, Arcfra positions Neutree around standardized model delivery, unified compute and model management, model governance, gateway management and end-to-end observability.
Paired with the Arcfra Enterprise Cloud Platform, the company says enterprises can build production-grade, governable model inference infrastructure. The unified model gateway and model registry are designed to help enterprises manage AI models across their lifecycle, from validation to production deployment. What to watch next: whether Arcfra names the manufacturing and shipbuilding customers behind those early deployments, and how quickly the Flex Engine’s non-LLM model support draws in teams whose document parsing and OCR workloads have until now sat outside their language model stack.
No pricing or availability details beyond the release date were provided.
