XtalPi

Genuinely hybrid rather than model led, and the company's own revenue disclosure is the clearest evidence for that grade. In 2024 the intelligent robot business contributed roughly 163 million yuan against roughly 104 million yuan from drug discovery, meaning the larger revenue line is automation hardware rather than AI driven discovery.

The computational stack is also physics anchored in the same way Schrodinger's is, built on quantum physics first principles calculations and the XFEP free energy perturbation engine, which run without machine learning. Generative AI and a multi agent orchestration system are load bearing in the Design Make Test Analyze loop and are not decorative, which is why this sits at B rather than C, but a buyer should understand they are purchasing an integrated physics, AI and robotics system in which the models are one of three legs.

Real autonomy at the execution layer. A multi agent system orchestrates large scale automated chemical synthesis and drives Design Make Test Analyze cycles, with the robotic laboratory explicitly positioned as closing the gap between computational design and wet lab validation, so predictions are physically tested rather than accumulating unchecked. The objects being automated are laboratory operations, not decisions affecting a patient.

Held at B because no confidence thresholds, human approval gates or published failure rates were located, so the boundary between what the multi agent system decides and what a chemist decides is described rather than specified.

Components are named freely but performance is never quantified in public. Disclosed elements include quantum physics first principles calculations, the proprietary XFEP platform for binding affinity prediction, generative AI for molecular design, and a multi agent system for synthesis orchestration, which is a specific architecture rather than a vague AI claim.

What is missing is any number a reader could evaluate: announcements describe breakthrough hit rates and enhanced accuracy without stating what the hit rate was, what the baseline was, or how accuracy was measured, and no peer reviewed methods paper, published benchmark or model documentation was located in this review. For a company listed on a public exchange the gap between commercial disclosure, which is strong, and technical disclosure, which is not, is notable.

The clinical form of this axis does not reach this product and is graded on the analogous one rather than marked not applicable. The platform operates on molecular structures, computational chemistry output and experimental assay data, with no patient records in the workflow, which bounds the chain as a matter of what the product does.

The architecture is also named at component level, covering first principles calculation, a binding affinity platform, generative design and synthesis orchestration, so a partner knows what stages their material passes through even if not who operates them. The confidentiality obligation covering partners' undisclosed targets and compound data sits in agreements rather than in published terms.

What is absent is enumeration: no hosting arrangement, no sub processor list, no retention position, and no statement of whether partner submitted structures or assay results improve models used for others. The multi agent synthesis orchestration adds a question specific to this architecture, since orchestration implies components that may be externally supplied and a partner should know whether any stage of their chemistry passes through a third party service. Ask which components are built rather than licensed, who hosts them, for a sub processor list, and for the training position on partner submitted material.

A platform derived molecule is in humans, which is the evidence class that matters most here. PEP08, a PRMT5 inhibitor discovered through the platform for partner PharmaEngine, received clinical clearance in June 2025 and reached a Phase I enrollment milestone in solid tumors in May 2026, with the partnership subsequently expanding into a second synthetic lethality programme, which is a partner voting with its budget after seeing the first result.

Operational scale is also real and independently checkable through exchange filings, with more than 300 enterprise and research clients and disclosed revenue growth of 53 percent in 2024. Held at B because no efficacy data exists yet, the platform performance claims underpinning the deals are unquantified as noted on the transparency axis, and the largest disclosed agreement by value remains at the beginning of its work rather than at an outcome.

Not applicable in the provider sense and rated accordingly rather than penalized. The platform operates on molecular structures, computational chemistry output and experimental assay data, with no patient records in the workflow. The confidentiality obligation covers partners' undisclosed targets and compound data.

Not applicable. Counterparties are pharmaceutical, materials, agricultural and energy R&D organizations rather than covered entities transferring protected health information.

No SOC 2, ISO 27001 or equivalent attestation was located and no trust centre was found. Two searches were run and a retrieval caution belongs on this row: financial data sites carrying this company's stock page publish their own compliance credentials on the same page, and those belong to the data provider rather than to this vendor. Do not credit them.

One consideration is specific to this company and should be raised neutrally rather than avoided. It is headquartered in Shenzhen with a Boston presence and serves Western pharmaceutical partners who place undisclosed target selections and proprietary chemistry into its environment, so cross border data handling is a diligence question here that does not arise in the same form for single jurisdiction peers.

There is, however, a specific place to look that the company's product pages do not advertise. It is listed in Hong Kong and publishes a full annual report containing both a corporate governance report and an environmental, social and governance report, alongside interim reports, all through its investor relations site. Under the Hong Kong listing rules that sustainability disclosure is mandatory and its prescribed scope includes customer data protection and privacy, so the company is obliged to say something about how it handles customer information in a document it signs off on. That is a materially better source than marketing material, and it was not retrieved here.

Graded C because nothing is published on any surface a buyer would ordinarily check and the report content was not obtained in this review. A diligence process should read the data protection section of the most recent annual report before concluding the company has said nothing, and should then ask whether that description extends to the partner chemistry held on the discovery side of the business rather than only to employee and customer personal data.

Correctly structured, with the platform not presented as a regulated device and regulatory standing sitting at the asset level through partners. The important precision for buyers is geographic, the same trap this index flags elsewhere: PEP08 received clinical clearance in June 2025 and is in Phase I evaluation for solid tumors in Australia and the Taiwan region, and no FDA specific clearance, US IND acceptance or US registered trial for a platform derived asset was located in this review. Clinical stage status in one jurisdiction should not be read as regulatory standing in another.

A publications search overturns the earlier absence finding on one half of this axis while leaving the other half exactly where it was, and the split runs along the company's two business lines rather than across it.

On technical validity the disclosure is among the strongest in this category. The company participates in the independent blind tests of crystal structure prediction organised by the Cambridge Crystallographic Data Centre, with its United States and China entities both appearing among the contributing groups in the seventh test. A blind test is the most demanding form of external evaluation available: an independent body withholds the answers, participants submit predictions, and the results are published whether a method performs well or badly. It has also published its prediction method in a peer reviewed journal with explicit large scale validation and a blind study, and separately published a protocol comparison quantifying the accuracy cost tradeoff, stating plainly that its faster protocols reduce wall time at the expense of force field accuracy in structure generation and ranking while still recovering most low energy structures. Naming where a cheaper configuration degrades is precisely what this axis asks for.

What remains unanswered is a different kind of governance question and it is the one with commercial teeth. The company states that its platform accumulates standardised experimental and computational data from real world projects in a closed feedback loop that continuously optimises the models, which means partner derived experimental data improves models that subsequently serve other partners. No disclosure was located on what boundaries apply to that flow, what ring fencing exists, or what happens to a trained model at termination. The same question applies to Genesis Therapeutics in this category and a buyer should put it to both in writing.

Worth noting for anyone reading across this record: the solid state prediction business publishes method papers and enters blind tests, while the newer discovery side names components without quantifying performance. One company, two disclosure cultures, and a single grade can hide that.

Components are named freely and performance is never quantified in public, which produces an unusual gap for a company of this kind. The disclosed architecture is specific rather than atmospheric, covering first principles physical calculation, a named binding affinity platform, generative design and a multi agent system for synthesis orchestration, so a reader knows what the pieces are. What is missing is any number they could evaluate.

Announcements describe breakthrough hit rates and enhanced accuracy without stating what the hit rate was, what baseline it improved on, or how accuracy was measured, which is the claim shape this index records as constructed so it cannot fail: an improvement with no denominator and no comparator cannot be wrong. No peer reviewed methods paper, published benchmark or model documentation was located, and no warranty, indemnity or remediation commitment.

The contrast with the company's own practice elsewhere is what makes this notable rather than routine. As a listed company it discloses commercially to a standard set by an exchange, with audited figures, stated methodologies and consequences for misstatement, and its technical disclosure meets none of that. The same organisation is therefore capable of rigorous public reporting and applies it only where an external obligation exists. Ask for the hit rate with its denominator and baseline, the measurement method behind any accuracy claim, and a peer reviewed methods paper.

Not applicable. This is a preclinical discovery and materials platform with no provider workflow surface and no EHR touchpoint.

Unusual for this category in that something physical is delivered: the intelligent robotics business sells automation systems to enterprise and research clients and is the larger of the two revenue lines, so a customer can operate XtalPi built laboratory hardware on its own site. The computational side does not follow the same pattern.

Drug discovery work runs on the company's own high performance cloud computing and standardized automation infrastructure, no on premise or private cloud option for the modelling stack was located, and no tenancy or regional data residency terms were published. Graded C rather than Not Rated because the axis genuinely applies and the disclosure is thin against a cross border profile that makes it matter more than usual.

The most granular commercial disclosure in this category, because a Hong Kong Stock Exchange listing forces it and the company was the first to list under the Chapter 18C framework in February 2025. Filings and reporting expose figures private peers never publish, including 2024 revenue of 266 million yuan up 53 percent with accelerated second half growth of 73 percent, the segment split between roughly 163 million yuan from intelligent robotics and roughly 104 million yuan from drug discovery, and an adjusted net loss narrowing 12.5 percent to 457 million yuan.

Deal disclosure is equally specific, including the DoveTree Medicines collaboration at up to 5.99 billion dollars with a second payment of 19 million dollars confirmed as received, and a June 2026 GPCR programme at over 400 million dollars total potential value. Publishing the revenue mix is the disclosure that most changes how a buyer reads this company.

Among the broadest in this category. Within therapeutics the platform covers small molecules and, through the Ailux biologics subsidiary, bispecific antibodies with Eli Lilly, spanning oncology, immunology, inflammation, neurology and metabolic disease in the disclosed DoveTree scope. Beyond therapeutics the same infrastructure serves materials discovery across agriculture, energy and cosmetics, reaching more than 300 enterprise and research clients in total.

Buyers should note that breadth this wide across regulated and unregulated industries means the depth of validation differs sharply by application, and only the small molecule oncology path has produced a clinical stage asset.