Insilico Medicine
Insilico Medicine is a generative AI drug discovery company (HKEX: 3696), headquartered in Cambridge, Massachusetts with operations in Hong Kong, Abu Dhabi and Shanghai, that both licenses software and develops its own medicines. Its software suite covers the discovery chain: PandaOmics for target identification and indication prioritization, Chemistry42 for generative molecular design using generative tensorial reinforcement learning rather than library screening, Generative Biologics, inClinico for clinical trial outcome prediction, and Science42 DORA for research writing, alongside language model work the company publishes as Nach01, PreciousGPT and its MMAI Gym training environment. A PandaOmics Box is offered for local deployment, and client work runs on the client's preferred cloud under an ISO 27001 certified information security management system.
The internal pipeline runs to more than 40 programs, 34 preclinical candidates nominated since 2021 and 13 cleared for IND, across fibrosis, oncology, immunology and metabolic disease. Its flagship is rentosertib, a TNIK inhibitor for idiopathic pulmonary fibrosis whose target the platform selected and whose molecule it generated: discovery was published in Nature Biotechnology, Phase IIa results in Nature Medicine, and its Phase III trial dosed its first patient in September 2026. The company also publishes public benchmarks for drug discovery and science AI.
Capability Axes
An AI Health Index grade measures what a buyer can verify from public sources on the date shown. It is not a rating of how good the product is. A vendor can build an excellent system and grade low on an axis because it publishes nothing an outsider can check. How grades read
Generative models run both sides of the business. On the software side, research organizations license generative tools: PandaOmics for target identification and indication prioritization, Chemistry42 for generative molecular design, inClinico for trial outcome prediction, Generative Biologics, Science42 DORA for research writing, and language model work the company groups under Nach01, PreciousGPT and its Large Language of Life Models.
The internal pipeline is the same models pointed at the company's own programs: more than 40 programs with 34 preclinical candidates nominated since 2021 and 13 cleared for IND, led by a TNIK inhibitor the company says its models selected the target for, designed the molecule for and informed the trial design of.
Discovery outputs are checked downstream. Target rankings are hypotheses a research team selects from, generated molecules are synthesized and tested before they count, and on the flagship program a platform selected target and generated molecule went through preclinical work, IND clearance and human trials now at Phase III, published at two stages. inClinico, the trial outcome module, works differently: its predictions inform whether a program proceeds, and a predicted failure that stops a program is never tested by a trial.
What uncertainty a user sees on a ranking or a prediction, and what the system declines to answer, is not published.
Insilico names its engines and the methods inside them. Chemistry42 generates molecules with generative tensorial reinforcement learning, the modules PandaOmics, Chemistry42, inClinico, Generative Biologics and Science42 DORA are each described separately, and the medicinal chemistry that came out of them is in the peer reviewed literature. The company also publishes language model work under its own names, including Nach01 and PreciousGPT and a training environment it calls MMAI Gym, runs public benchmarks on drug discovery and science tasks, and announces platform updates by quarter.
No model card, technical documentation or version identifier is published for any module.
Every model in the suite is Insilico's own product, and each client's work runs on infrastructure the client chooses, AWS, Azure or another provider, with project by project access controls, encryption and data destruction at the end of an engagement.
Whether any outside model provider sits inside those models, or whether one client's material informs models another client uses, is not published.
Rentosertib, a TNIK inhibitor for idiopathic pulmonary fibrosis whose target the platform selected and whose molecule it generated, had its discovery published in Nature Biotechnology and its Phase IIa results in Nature Medicine, reporting a mean forced vital capacity change of +98.4 mL at 12 weeks in the 60 mg arm against -20.3 mL for placebo, and entered Phase III in September 2026, when the trial dosed its first patient. Thirteen programs have reached IND clearance and 34 preclinical candidates have been nominated since 2021, and the pipeline is listed program by program, with target, indication and development stage. Platform module work is published separately, including a PandaOmics study in npj Precision Oncology in 2026.
The company puts its platform speed at 12 to 18 months to a preclinical candidate against an industry norm of 2.5 to 4 years, with 60 to 200 molecules synthesized per program, figures given without a matched comparison. No product is approved.
Client data is destroyed at the end of each engagement through controlled removal and decommissioning, with project by project access controls limiting who can see it in the meantime, encryption in transit and at rest, and research records held in an internal electronic lab notebook under confidentiality and intellectual property handling procedures.
Whether a client's data, targets or results train or improve the models other clients use is not stated. The company's published privacy policy (effective October 30, 2020) covers the company website rather than the software.
The company's target identification work runs on human omics and clinical datasets, and its inClinico product predicts trial outcomes. Its only published privacy policy (effective October 30, 2020) covers the company website and the information exchanged with it, and its security controls govern how client data is handled and destroyed.
Whether the platform may receive protected health information, and under what agreement, is not stated.
An ISO 27001 certified information security management system governs the company's processes. Around it sit project by project access controls restricting data visibility to authorized personnel, encrypted communications and secure environments for data in transit and at rest, redundant backups, continuous monitoring and regular vulnerability assessments, and controlled data removal and decommissioning so that client data is destroyed at the end of an engagement. Research records sit in an internal electronic lab notebook and controlled document repositories, with confidentiality and intellectual property handling procedures. The company operates from Boston, Hong Kong, Abu Dhabi and Shanghai.
No certificate, scope statement, certifying body or expiry date is published for the certification.
Research teams use the suite for target identification, molecular design and trial outcome prediction, and the medicines it helps design carry their own regulatory status. Several of those medicines have reached the clinic: the company reports 13 programs cleared for IND, and rentosertib, its TNIK inhibitor for idiopathic pulmonary fibrosis, followed Phase IIa with a Phase III trial that dosed its first patient in September 2026.
Nothing presents the software itself as a regulated device.
The platform modules and the flagship program appear in peer reviewed venues, and the company publishes benchmark efforts for the field, including a drug discovery benchmark and a science benchmark on its own domains, which evaluate methods in general rather than its own models. Target identification mines published biomedical literature and public omics data, a body of work concentrated on a small subset of well studied genes, and the company points to a novel target it found for a fibrotic disease as an answer to that concentration.
No governance framework, model card, statement of intended use or evaluation of where its own models fall short is published.
A ranked target is a hypothesis a research team selects from, a generated molecule has to be synthesized and tested, and on the company's own programs the chain continued into preclinical work, IND clearance and controlled clinical trials, with two stages published. Every consequential output is checked before it counts.
No accuracy, hit rate, warranty or remediation commitment is published for PandaOmics, Chemistry42 or inClinico.
The discovery software serves research teams, so the record systems that matter to its buyers are electronic lab notebooks, laboratory information systems and research data platforms. Insilico describes a generic integration path: the suite runs on the client's preferred cloud (AWS, Azure or others as specified), a PandaOmics Box is offered for local deployment, and the company has published work on connecting its tools to other software through standard model context interfaces. The electronic lab notebook Insilico describes among its security controls is an internal system.
No connection to a customer's research systems is named.
Each client engagement runs on the client's preferred cloud servers, whether AWS, Azure or another provider the client specifies, so that regional data residency and sovereignty requirements can be met. Access controls are set project by project to keep one engagement's data apart from another's, alongside encrypted transport and storage, redundant backups, and controlled removal and decommissioning at the end of an engagement. A PandaOmics Box is offered alongside the hosted suite.
The company operates from Boston, Hong Kong, Abu Dhabi and Shanghai, and which of those locations' personnel or compute touch a given engagement is not stated.
Insilico sells software licenses to research organizations, runs collaborations and licenses its own assets out; its published pipeline names the partners holding rights to out licensed programs, among them Exelixis, Menarini and TaiGen, and offers further programs in discovery for licensing through a business development address. Each product page routes to a request form, a PandaOmics Box is offered alongside the hosted products, and as a listed company Insilico reports its own financial results, which show its revenue rather than a customer's cost.
No tier, unit of charge or rate card is published.
The platform spans target identification, molecular generation, biologics and trial outcome prediction, and the company's own pipeline covers fibrosis, oncology, immunology and metabolic disease, publishing each program's target and indication. Results sit where the programs have run: a fibrosis candidate in Phase III, thirteen programs cleared for IND, and published work behind them. The company also states extension beyond human therapeutics into materials, agriculture, nutritional products and veterinary medicine, and those are named without a customer or a result attached.
Insilico licenses discovery software while developing its own medicines in overlapping indications, so a customer's targets and chemical series sit with a company running competing programs.
Nothing published sets out which therapeutic areas its own programs exclude, or what separates the two sides.
Product updates
Material product, regulatory, evidence and commercial changes at Insilico Medicine, each verified against a live source and tagged to the capability axis it bears on. Funding rounds and awards are not product changes and are not logged.
Insilico announced that Model Context Protocol servers are now available across its Pharma.AI software, so outside AI agents can connect directly to its biology, chemistry and biologics engines, including Generative Biologics, PandaOmics and Chemistry42, and run multistep discovery workflows. The announcement previews a 30 September webinar that will present new PandaOmics Agent skills and other platform updates.
Insilico released a set of small language models trained as scientific specialists for chemistry and biology through its MMAI Gym for Science framework. The set covers chemical synthesis, ADMET prediction and potency prediction across GPCR and kinase panels, and includes a single step retrosynthesis model built on Liquid AI's 2.6 billion parameter architecture.
Insilico Medicine convened the Open Consortium for Benchmark Quality in AI Driven Drug Discovery, known as O3DC, and published a live catalog of the benchmarks the field uses to claim performance. The catalog tracks each benchmark's metrics, maintainers and repository activity, and unusually it documents the known caveats, biases and limitations of each one rather than presenting them as neutral yardsticks. The stated premise is that a benchmark with an undisclosed bias produces model comparisons that look rigorous and are not.
Alternatives to Insilico Medicine
Vendors in the same categories as Insilico Medicine whose graded profiles look most alike, with a line on what each product does.
Similar is not the same, so check that each one does the job you need. No vendor pays for placement.
Edison Scientific is the commercial spinout of the nonprofit research lab FutureHouse, formed in November 2025 to run its AI scientist platform.
BigHat Biosciences, founded in San Mateo, California in 2019 by Mark DePristo and Peyton Greenside, designs and engineers antibodies with machine learning tied to its own automated…
Iktos is a Paris generative AI and robotics company for small molecule drug discovery, founded in 2016.
Tech bio company building what it calls an AI scientist for biopharma research and development, combining biological reasoning models, a multimodal patient data network and agentic software.
Protein engineering software company, based in Amsterdam and Zurich, that licenses an AI platform directly to R&D teams rather than working through discovery partnerships.
Clinical stage generative biology company in Somerville, Massachusetts (Nasdaq: GENB), founded in 2018 by Flagship Pioneering, that designs protein therapeutics with machine learning rather…
Ordered by how closely the graded profiles match across the fifteen axes, inside the same categories. See how each axis is graded in the methodology.
Compared With
Each comparison carries a written verdict, the buyer conditions that favor each vendor, and a graded side by side. Pairs that cross a category boundary are grouped separately, and their verdicts state where the boundary sits rather than manufacturing a head to head.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | BAA Tier | Implementation | Source |
|---|---|---|---|---|
|
Not published
|
Software licenses for PandaOmics, Chemistry42, inClinico and the rest of the discovery suite, sold through contact, plus discovery collaborations and out licensing of the company's own assets. No tier, unit of charge or rate card published. | — | Not published. A PandaOmics Box is offered for local deployment alongside the hosted products, with no price attached. | Vendor Published |
The business combines software licenses for research organizations, discovery collaborations and out licensing of its own assets, and further programs in discovery are offered to partners through a business development address. Every product page in the suite routes to a request form, and the PandaOmics Box is offered for local deployment. As a company listed in Hong Kong, Insilico reports its own financial results, which describe what it earns rather than what a customer pays. No price, tier or rate card is published.