Funding
INFevo Raises Over RMB 100M Pre-A to Expand AI Virtual-Cell Research

Key PointsINFevo pairs virtual-cell prediction with automated wet-lab validation to close the…
View more
- INFevo pairs virtual-cell prediction with automated wet-lab validation to close the loop between hypothesis and experiment.
- Karl Workbench's top SciAgentArena ranking covers only DeepSeek in two omics categories rather than broad scientific leadership.
- The decisive test is whether benchmark results translate into repeatable performance on patient-derived organoids and pharma workflows.
INFevo announced on November 16, 2026 that it had raised more than RMB 100 million in Pre-A funding to expand its AI virtual-cell models and automated experimental systems, according to investor Xinglian Capital’s announcement. The proceeds will also support deeper collaboration with pharmaceutical companies, hospitals and research institutions.
Xinglian Capital, Prosperity7 Ventures, Huagai Capital, Huakong Fund and DT Capital joined the round. Existing investors Shunwei Capital, HSG and Songhe Capital made additional investments. The financing followed an angel round of tens of millions of yuan in June 2026, making it INFevo’s second in five months; Xinglian described it as a record single-round financing for a Chinese virtual-cell company.
Pairing Virtual Cells with Automated Experiments
INFevo’s virtual-cell approach is designed to predict cellular responses to drug or genetic interventions before laboratory experiments. Researchers could then compare potential targets, rank candidate molecules and reserve wet-lab capacity for more promising questions. XtalPi, which incubated INFevo, described in a July 16, 2026 announcement how its molecular-design and automated-synthesis capabilities could complement those predictions, creating a chain from molecular generation through cellular-response prediction to experimental validation.
The OCOO-T paper describes a minimalist flow-matching model whose standard Transformer stack operates directly on continuous gene-expression profiles. Prediction is formulated as continuous-time denoising, while perturbation, dosage and cell-line or cell-type information enter through adaptive layer normalization and in-context tokens.
The authors report state-of-the-art results across the Tahoe100M, Replogle and PBMC benchmarks, covering chemical, genetic and cytokine perturbations. The model also scales to long transcriptional profiles by patching and depatching cellular contexts.
Beyond a single model, INFevo is building what it calls recursive scientific discovery: retaining scientific questions, hypotheses, analytical methods and experimental results so they can inform later hypotheses, experiment designs and candidate selection. Its automated wet-lab system covers cell culture, perturbation and multi-omics measurement. Model predictions help choose experiments, while the resulting data test those predictions and feed subsequent model improvements.
Research Agents and Their Benchmark Limits
The Popper Project addresses the hypothesis side of that research process. Named after philosopher Karl Popper, it uses research agents and professional databases to generate, challenge and refine testable hypotheses, displaying traceable reasoning chains, evidence gaps and proposed experiments. Its Karl Workbench has entered beta testing, with research records, accumulated evidence, outputs, supervision and iterative-improvement functions that let researchers review judgments and resume earlier tasks.
Xinglian reported that an October 2026 retest using SciAgentArena placed Karl Workbench first overall in both single-cell and spatial omics when running on DeepSeek. The benchmark comprises approximately 200 stepwise-verified tasks across drug discovery, single-cell omics, spatial omics, electronic health records and genetics. It separates agent execution from evaluation and assesses data analysis, optimization, discovery and validity checking.
SciAgentArena’s authors found that no single agent dominated across tasks and domains. Current systems performed unevenly, had difficulty with novel discovery and self-directed exploration, and tended to execute requests even when the scientific premise was invalid. That wider record makes the scope of INFevo’s reported result important: the retest concerns DeepSeek in two omics categories, rather than a general lead across scientific research.
Clinical Samples Provide the Next Test
INFevo is also participating in a Beijing key laboratory led by Peking University First Hospital that focuses on liver, biliary and pancreatic tumor organoid culture and intelligent diagnosis and treatment. The collaboration’s stated aim is to connect clinical questions, patient-derived organoid models, drug screening, AI analysis and clinical validation. Model predictions and organoid experiments are intended to inform one another, bringing the virtual-cell system into contact with patient-derived clinical samples.
Co-founder Yalong Zhao said the financing would support the foundation-model and automated-experiment infrastructure and accelerate the recursive-discovery system. Lipeng Lai, another INFevo co-founder, is also XtalPi’s co-founder and chief innovation officer.
The new capital gives INFevo resources to expand both sides of that system: computational prediction and automated validation. The consequential test is whether developer-reported benchmark results and beta research tools can become repeatable performance on patient-derived samples and in pharmaceutical research workflows.












