SyntheticGestalt launches molecular AI models for drug discovery

SyntheticGestalt on July 21, 2026 launched two molecular AI models, ZAO and KOYA, for drug discovery and chemistry and made both available through the NVIDIA BioNeMo Agent Toolkit. The company says the tools are designed to help AI agents and researchers speed molecule screening, design, and optimization in shared workflows.

Why it matters: - SyntheticGestalt is pushing molecular AI deeper into early-stage drug discovery, where faster screening and better molecule design can shorten the path from idea to lead candidate. - The NVIDIA BioNeMo Agent Toolkit integration means AI agents can use ZAO and KOYA alongside other life-science tools in a single workflow. - The models are offered for enterprise use, with pricing and terms set individually.

What happened: - SyntheticGestalt announced two molecular AI foundation models on July 21, 2026: ZAO and KOYA. - ZAO is a molecular foundation model for small-molecule information. - KOYA is a molecular generative model for designing novel ligands for target proteins. - Both models can now be called from the NVIDIA BioNeMo Agent Toolkit, which NVIDIA announced in June 2026. - The announcement came from SyntheticGestalt in Tokyo.

The details: - ZAO converts the chemical structure of small molecules into data AI can use. - ZAO uses a "4D" representation that handles multiple 3D conformations a molecule can adopt. - SyntheticGestalt says that approach captures three-dimensional shape and flexibility better than flat structures or a single conformation. - ZAO was pre-trained on about 10 billion compounds. - SyntheticGestalt says ZAO can build high-accuracy predictive models from as few as several hundred data points. - ZAO produces stable, reproducible features because its output does not depend on which conformation is selected. - Without task-specific fine-tuning, ZAO ranked No. 1 on nine public benchmark tasks, including ADMET and activity prediction. - SyntheticGestalt says that result is the world's best performance for a single general-purpose model. - KOYA is built to design new molecules while closely reflecting medicinal chemists' intent. - KOYA supports precise design control, including fixing a scaffold and growing a molecule in a specific direction. - KOYA preserves target-protein interactions that chemists want to keep. - KOYA lets chemists set acceptable element choices and molecular-size constraints. - KOYA removes hard-to-synthesize or unrealistic structures before presenting candidates. - KOYA ranks outputs using drug-likeness and ligand efficiency. - Across 10 evaluated targets, KOYA generated novel molecules that outperformed the known binding ligand in docking evaluation on 8 targets. - The BioNeMo Agent Toolkit offers standardized skills such as structure prediction, molecular generation and docking. - ZAO and KOYA can be used through an agent to convert molecules into data and design new ligands in the same workflow as other life-science tools. - SyntheticGestalt says the models build on its work in large-scale compound exploration, including a 38-billion-compound virtual screening with Enamine. - SyntheticGestalt is also advancing proof-of-concept work with major domestic pharmaceutical and chemical companies.

Between the lines: - The launch positions SyntheticGestalt to compete on both ends of the discovery workflow: evaluating compounds and generating new ones. - BioNeMo compatibility could make the models easier to adopt inside agent-based R&D systems already under development. - The company is packaging prior research into commercial products, suggesting a shift from demonstrations to enterprise deployment. - CEO Koki Shimada framed the launch as part of an emerging era in which AI agents can independently support research and work with scientists.

What's next: - SyntheticGestalt will sell ZAO and KOYA as enterprise products with negotiated pricing and terms. - The company is likely to push adoption through agent-driven workflows in drug discovery and chemistry. - SyntheticGestalt will continue proof-of-concept work with pharmaceutical and chemical companies as it expands the product line.

The bottom line: - SyntheticGestalt is trying to turn molecular AI from a research asset into a commercial drug-discovery workflow tool, with NVIDIA integration as a distribution and adoption boost.

More information: SyntheticGestalt's website