Code Engraving | Enhe Technology Rebrands and Launches SAION AI: The World's First Physical AI Platform for Bio-manufacturing

On March 9, 2026, MaHui portfolio company Enhe Technology officially announced the launch of SAION AI, a Physical AI platform for the bio-manufacturing sector. At the same time, the company announced that its Chinese abbreviation would change from "Enhe Biotechnology" to "Enhe Technology." This name change represents more than a brand upgrade — it signals a clear strategic pivot, from a biotech-centric enterprise toward a technology company that drives industrial innovation through intelligent systems.

Since its founding in 2019, Enhe has consistently focused on one core question: how to advance bio-manufacturing from experience-dependent experimental exploration toward a scalable, intelligent engineering system. With the deep integration of artificial intelligence, automation, and data systems into bio-manufacturing, the company has gradually built data-driven R&D and production infrastructure. The launch of SAION AI marks Enhe Technology's formal entry into a new phase of Physical AI-driven bio-manufacturing.

SAION AI: A Physical AI Platform for Bio-Manufacturing

01

SAION AI: Bringing AI Into the Physical World of Experiments

While artificial intelligence is already reshaping the digital world with its powerful cognitive and generative capabilities, another more transformative force has entered the physical realm — Physical AI, an intelligent system capable of perceiving, understanding, and engaging with real physical environments, directly participating in task execution and making rational decisions. Today, Enhe Technology officially launches SAION AI, the world's first Physical AI platform purpose-built for bio-manufacturing.

SAION AI is not merely an AI agent for virtual design or a single-function lab automation tool. It is a Physical AI platform encompassing cognition, control, and closed-loop execution capabilities, capable of autonomous design, direct participation, and optimization of biological discovery and production processes. It generates executable experimental plans based on scientific intent, communicates directly with bio-foundries through Enhe's self-developed Biology Protocol Language (BPL), standardizes the completion of real-world experiments, and continuously evolves through data feedback loops — addressing the industry's challenges of lengthy R&D and production chains, numerous procedures, fragmented data, and heavy reliance on human experience and trial-and-error.

02

Platform Architecture: Cognition, Control, and Closed-Loop Execution

In its architectural design, SAION AI adopts Physical AI as its core philosophy, building a collaborative evolution architecture (COE Model) composed of the Cognition Layer — Orchestration Layer — Closed-loop Execution Layer. This architecture can be analogized to the Vision-Language-Action (VLA) models that have attracted significant attention in the Physical AI field. VLA models establish unified architecture and cognitive reasoning capabilities under native multimodal large models, breaking away from traditional modular and rule-driven paradigms and enabling efficient data evolution and the emergence of intelligent application scenarios.

Through its self-developed three-layer architecture, SAION AI achieves unified internal scheduling and coordination, allowing it to operate in complex, long-chain bio-manufacturing industrial scenarios. It leverages multi-scale deep understanding of living systems in the digital dimension, intelligent task orchestration and tool scheduling, and extends to physical task execution and data feedback — forming a self-optimizing intelligent closed loop within the platform.

Cognition Layer: Multi-Scale Understanding of Living Systems

The Cognition Layer is built upon the long-term data accumulation from Enhe's self-developed Cell2Cloud bio-foundry, integrating tens of millions of real project closed-loop experimental data points, millions of literature sources and patents, and incorporating specialized biological databases including NCBI, UniProt, and PubMed.

The system integrates multiple AI4Science models including AlphaFold, ProteinMPNN, RFDiffusion, and ESMFold, covering capabilities in protein structure prediction, sequence generation, metabolic pathway analysis, enzyme engineering, and fermentation data modeling. This enables SAION AI to systematically comprehend living systems across scales from genes to proteins to metabolism to cells to fermentation, identifying optimal R&D directions within vast design spaces and providing cross-scale contextual data foundations for subsequent scientific decision-making.

Orchestration Layer: Dynamic Coordination Hub

The core of the Orchestration Layer is the Agent Harness intelligent agent orchestration engine, which uses large language model reasoning to uniformly coordinate multi-agent collaboration, tool invocation, and task execution. The system can parse complex scientific objectives into structured task graphs and build Workflow Skills based on the company's accumulated experience in strain development and bio-manufacturing, forming stable scientific execution patterns.

The platform has already integrated 316 specialized scientific research tools, achieving dynamic composition of model and algorithm capabilities through intelligent tool routing. Checkpoint and fault-tolerance mechanisms support stable execution of long-duration, complex scientific workflows, constituting SAION AI's central hub for scientific decision-making and task scheduling.

Execution Layer: Standardized Experiment Execution and Data Closed Loop

The Execution Layer uses Enhe's self-developed Biology Protocol Language — BPL — to translate experimental plans generated by SAION AI into standardized experimental instructions that directly drive equipment execution, achieving automated flow from R&D planning to experimental operations.

The system interfaces with Biofoundry APIs to intelligently schedule liquid handling workstations, cultivation, and detection equipment, while monitoring experimental progress and equipment status in real time. Experimental data is automatically parsed and structurally fed back into the platform, driving continuous model optimization through reinforcement learning and forming a Design–Build–Test–Learn (DBTL) closed loop — continuously strengthening SAION AI's scientific capabilities and accumulating knowledge assets.

Through this architecture, SAION AI deeply integrates scientific cognition, intelligent decision-making, and physical experiment execution, comprehensively constructing an AI-driven closed-loop system for bio-manufacturing.

03

Real-World Performance

SAION AI has achieved state-of-the-art (SOTA) performance on multiple international life science AI benchmarks, systematically validating its core scientific capabilities as an AI Scientist. In key research tasks including literature comprehension, biological sequence reasoning, genetic engineering design, and scientific discovery, SAION significantly outperforms general-purpose large models and multiple specialized models.

• Scientific Literature Comprehension: Achieved 70.7% average accuracy on LitQA (Lab-Bench) and SuppQA (Lab-Bench) benchmarks, significantly outperforming current mainstream foundation models (GPT-5.3, Opus 4.6) by nearly 20 percentage points, as well as the publicly reported results for the research-optimized model Stella (LitQA 65.0%).
• Biological Sequence Analysis: Reached 88.2% accuracy on SeqQA (Lab-Bench), outperforming current mainstream foundation models and exceeding the Biomni platform (81.9%) reported in Stanford University literature, demonstrating leading DNA/RNA/protein sequence reasoning and design capabilities.
• Genetic Engineering and Experimental Design: Achieved 84.9% average accuracy on Gene Editing (Lab-Bench) and Cloning Scenarios (SDE) benchmarks, reaching SOTA levels among current models and validating its reasoning capabilities in real molecular biology experimental design.
• Scientific Discovery and Reasoning: Scored 89.6% accuracy on the BAIS-SD benchmark (evaluating whether agents can generate novel biological scientific discoveries and reasoning), an improvement of approximately 12 percentage points over mainstream benchmark models, reflecting leading capabilities in research hypothesis comprehension, scientific reasoning, and research discovery tasks.
• Real Experimental Validation: Since no existing benchmark can comprehensively evaluate AI models' closed-loop execution capabilities in biological experiments, we validated SAION AI's physical-level scientific performance through full-process real experiments. SAION has autonomously completed tasks from literature review to plasmid design to wet-lab assembly, achieving over 90% accuracy — proving that it not only excels in scientific comprehension and reasoning benchmarks but also possesses the capability to independently drive biological R&D in real experiments.

Across four core benchmarks and real experimental validation, SAION ranks first in multiple tasks. These results demonstrate that SAION has developed systematic capabilities spanning the biological research workflow — from scientific knowledge comprehension and sequence analysis to experimental design and scientific discovery — transforming AI from a knowledge tool into an AI Scientist model capable of driving real scientific work, significantly improving bio-manufacturing R&D efficiency and accelerating the translation of scientific discovery to the physical world.

04

Core Advantages and Characteristics

Based on the above architecture and technical achievements, SAION AI's core advantages and characteristics can be summarized in five points:

Dual-Source Knowledge-Driven Research Planning

SAION AI builds a cognitive model moat using tens of millions of proprietary experimental data points accumulated from real internal projects, combined with millions of public literature sources and patents. It autonomously combines and chain-invokes multiple cutting-edge specialized models, forming adaptive goal-oriented workflows that translate scientific intent into executable technical roadmaps, task plans, and solutions.

Experimental Task Plan Codification

Experimental task plans output by the SAION AI platform can be precisely translated through Enhe's self-developed Biology Protocol Language (BPL) into standardized experimental work orders for human operators and machine-executable instructions for equipment. As a standard protocol, BPL ensures reproducibility and traceability of experimental plans across different people, times, and equipment, guaranteeing compliance of experimental result data.

Asset-Aware Contextual Design Capability

Strains and biological components are core assets in bio-manufacturing. The SAION AI platform can automatically identify existing, reusable DNA fragments, standard plasmids, and strains in internal inventory during the experimental design phase, proactively recommending or automatically incorporating them into experimental plans. Simultaneously, during experimental execution, DNA designs, strain construction, transformation, and genetic information propagation results are automatically entered into databases, forming traceable strain construction pathways and complete strain physical status records.

Direct Drive and Intelligent Scheduling of Bio-Foundries

Through the BPL standardized protocol, SAION AI translates experimental plans into machine-readable instructions delivered directly to Enhe's self-developed Cell2Cloud bio-foundry for execution. This eliminates information transmission losses inherent in traditional biological experiments, improves experimental execution accuracy and reproducibility, and monitors experimental completion progress in real time. Furthermore, all experimental queues, equipment status, and consumables inventory within the Cell2Cloud bio-foundry are optimally and intelligently scheduled under SAION AI's drive.

Bio-Manufacturing-Specific Data Intelligence and Knowledge Accumulation

During task execution, SAION AI can autonomously acquire, track, and analyze result data in real time, supporting rational decision-making and enabling full-chain Physical AI intervention in bio-manufacturing processes. The accumulated proprietary data is stored as structured, queryable, and callable organizational data assets, empowering internal talent development and enabling precise design of experimental plans and process development — achieving continuous evolution at all levels of the SAION AI platform.

05

From R&D Closed Loop to Intelligent Manufacturing Closed Loop

The deep real-world application of Physical AI expands our ability to understand life and is transforming traditional experimental and production methods. The launch of SAION AI signals that bio-manufacturing is entering a continuous self-reinforcing modality of digital cognition, intelligent orchestration, and closed-loop execution — propelling the industry from experience-driven trial-and-error toward intelligent engineering characterized by digital-hardware interactive perception and iterative leaps. The efficiency frontier of bio-manufacturing is being redefined.

About Enhe Technology (Bota): Enhe Technology (Bota) is a global leader in Physical AI-driven bio-manufacturing, dedicated to transforming biotechnology into a reliable, scalable industrial productivity engine. By integrating artificial intelligence, synthetic biology, and industrialized end-to-end capabilities, Bota has opened a new paradigm in bio-manufacturing R&D and production. We have built the world's premier physically intelligent bio-foundry, covering the entire process from strain engineering, process development, and scaled production — translating complex biology into scalable solutions for food, nutrition, personal care, and more industries. Bota partners with global customers to provide greener, more efficient biological solutions, accelerate industry adoption, and collectively drive the transition toward a sustainable future.