Shanghai Scinetics Intelligent Technology Co., Ltd. (Scinetics) has closed a first funding round of nearly 50 million yuan. The round was led by Innoangel Sci-Tech Fund, with participation from Yijing Capital, Unity Ventures and Linge Capital. Lighthouse Capital served as strategic incubator and exclusive financial adviser.

According to the company, the proceeds will primarily fund iteration of its AI4S scientific foundation model, large-scale deployment of its automated experiment platform, and hiring for the core team.

Scinetics is an AI4S company focused on scientific discovery, founded by researchers including Zhang Zaixi, an assistant professor at the Hong Kong University of Science and Technology. Its technical approach combines an AI4S foundation model with physical AI, aiming to link scientific reasoning, experimental design, execution and data feedback into a scientific intelligence system that can keep learning and iterating.

From Single-Purpose Models to Scientific Agents

Before founding Scinetics, Zhang and his team had already built several specialized models for life sciences, including the molecular screening model MGSSL, the drug molecule generation model FLAG, the protein pocket design model PocketGen, and the RNA design model RNAGenesis.

Each model targeted a specific research task. But a complete discovery or drug design workflow usually requires coordinating multiple models, databases and experimental tools across several rounds of validation. Even a model that excels at one task struggles to cover the full research pipeline on its own.

That conclusion pushed the team to broaden its focus to scientific agents, releasing systems such as STELLA and BioClaw that use agents to connect specialized models, databases and bioinformatics tools.

Published papers show STELLA uses a multi-agent architecture, handling biomedical research tasks through a continuously updated library of reasoning templates and a tool system. The paper is currently available on a preprint platform, and the benchmark results it reports show STELLA outperforming contemporaneous comparison models on some biomedical question-answering and literature retrieval tasks.

Scinetics says STELLA has served more than 1,000 research users, while the BioClaw community has over 1,200 members and has received more than 500 co-creation applications. Those user figures and partnership claims come from the company itself and lack third-party verification.

Unifying Scientific Data With Scientific Tokens

Alongside its scientific agents, Scinetics is developing a foundation model built for scientific discovery.

Research spans many data types: DNA, RNA, protein structures, medical images, omics matrices and experimental readouts. Each typically has its own representation, forcing existing agents to convert repeatedly between text, code and specialized tools. That can lose information and limits how directly a model can understand a scientific problem.

The company's proposed answer is a technical approach it calls Scientific Token, which converts scientific data across modalities into basic units a model can process uniformly. The goal is to let models learn relationships between scientific objects directly, rather than relying on a language model to call external tools.

Scinetics plans to first build model capabilities around the central dogma of life sciences, covering the links from DNA and RNA through to proteins, function and phenotype, then apply them to tasks such as protein design, target discovery and drug development.

The approach is still in development. Whether the model can meaningfully raise design success rates on real research tasks will require standardized benchmarks, experimental replication and peer review.

Closing the Loop From Hypothesis to Experiment

Model outputs ultimately need experimental validation. Scinetics is building a sandbox environment made up of more than 200 types of lab instruments and training an Agentic-VLA vision-language-action model that lets agents control robotic arms to run experiments.

As the company envisions it, the system needs to break down experimental tasks, operate instruments, monitor processes and recover from anomalies. When conditions change or a run fails, the agent should identify the problem and adjust what it does next.

The team is also trying to capture intermediate and failure data that traditional research workflows tend to discard, such as synthesis yields, reaction rates and images from the experimental process. Structured, that data can feed back into model training, turning experimental results into input for the next round of iteration.

The intent is a closed loop of cognition, design, experiment, feedback and further learning, moving AI systems from suggesting research directions to taking part in running experiments.

STELLA Screens Candidate Targets for Leukemia

In one application case disclosed by Scinetics, a Chinese state key laboratory working on acute myeloid leukemia used a privately deployed version of STELLA to look for potential therapeutic targets.

The company says STELLA produced a list of candidate targets in roughly 10 minutes, with the top-ranked candidate lacking direct prior literature coverage. The lab then ran preliminary validation across multiple cell lines and got positive results.

The full experimental design, the name of the candidate target, sample sizes and validation data have not been released, and no peer-reviewed paper has appeared. For now the case counts as early-stage research progress disclosed by the company, not a drug development result or evidence of clinical efficacy.

Extending Beyond Life Sciences

Zhang Zaixi, Scinetics founder, CEO and CTO, completed his undergraduate studies at the School of the Gifted Young at the University of Science and Technology of China, then earned a PhD in computer science there, with joint training at Harvard Medical School and postdoctoral research at Princeton University. His public biography states he joined the Hong Kong University of Science and Technology as an assistant professor in 2026, working on AI for Science, scientific foundation models, agents and embodied experimental platforms.

Scinetics is starting in life sciences and plans to extend its technology into materials science, quantum physics, aerospace and chip design.

Data formats, lab equipment and evaluation standards differ sharply across scientific fields, however. Whether models and lab automation capabilities developed for life sciences can transfer cleanly to other disciplines remains an engineering and commercial question the company has yet to answer.

As generative AI moves into drug design, materials development and lab automation, competition in AI4S is shifting from standalone prediction models toward system platforms that connect models, data, agents and real experiments. Whether Scinetics can build a stable loop between its scientific foundation model and its physical experiment system will be the main thing to watch in its next phase.