Best AI Drug Discovery Platforms

Insilico has the strongest evidence that AI drug discovery works end to end. Its PandaOmics software picked TNIK as a new target for idiopathic pulmonary fibrosis (a scarring lung disease), and its Chemistry42 generative chemistry designed rentosertib to hit it. In a randomised Phase 2a trial of 71 patients, published in Nature Medicine in 2025, the 60 mg daily dose improved lung capacity (FVC) by 98.4 ml over 12 weeks while the placebo group declined by 20.3 ml. The trial was small and short, but the result was strong enough to take forward: a Phase 3 trial (NCT07687459) began recruiting in September 2026.

Insilico says it has more than 40 programmes and 13 IND approvals. It went public in Hong Kong in late 2025, and in March 2026 signed a deal with Eli Lilly reported at up to $2.75 billion. Unlike most rivals, it also sells its software to other companies.

Pick it if you want to partner with or license from the company with the most advanced AI-discovered drug. Skip it if you need a purely Western partner; much of its R&D runs in China.

Schrödinger is the most accessible serious platform on this list: any pharma or biotech can license its software. Its core strength is physics-based simulation, especially free energy perturbation (FEP), which predicts how tightly a molecule will bind its target, combined with machine learning to search huge chemical spaces. Software sales brought in $68.1 million in the first half of 2026, which shows how widely it is used.

Its own drug record is solid but modest. SGR-1505 (MALT1, for B-cell cancers) and SGR-3515 (Wee1/Myt1, for solid tumours) are in Phase 1. Its August 2026 filing says it will finish those trials but will not start new ones on its own, and will look for partners instead, part of a restructuring expected to save about $70 million. Its platform also helped design molecules at Nimbus Therapeutics, a company it co-founded, including a TYK2 inhibitor now developed by Takeda.

Pick it if you want to run AI-assisted discovery in-house with mature tools. Skip it if you want a company that will carry your drug to market; it has stepped back from late-stage development.

Recursion runs one of the largest biology data engines in the industry. Robots photograph millions of cells after genetic or chemical changes, and AI turns those images into maps of how genes and drugs relate. In 2024 it merged with Exscientia to add AI chemistry, then cut programmes to focus its spending.

Its clinical record is real but still early. REC-4881, a MEK1/2 inhibitor for familial adenomatous polyposis (an inherited condition that causes hundreds of gut polyps), is in Phase 2 with Orphan Drug and Fast Track status, and Recursion began FDA talks on a registrational path in 2026. Four more drugs are in Phase 1, and REC-7735 has IND clearance; Recursion says it went from first hit to candidate in 10 months and 242 compounds. Some earlier programmes were dropped.

It scores well on openness because it releases models and data, such as the Boltz-2 structure and binding model built with MIT, and partners include Sanofi and Roche.

Pick it if you want open tools or a data-rich partner for target discovery. Skip it if you need late-stage clinical proof today.

Generate:Biomedicines, a Flagship Pioneering company, uses generative AI to design proteins, mainly antibodies. Its lead drug, GB-0895, blocks TSLP, a signal that drives airway inflammation, and is engineered to be injected only twice a year. Two Phase 3 trials in severe asthma started in December 2025 and January 2026, which makes it one of the most advanced AI-designed biologics anywhere.

It is worth being precise: TSLP is a proven target (an approved antibody already hits it), so the AI's contribution is a better-engineered molecule rather than a new biology discovery. That is still a meaningful test, because longer dosing intervals are a real advantage for patients.

The company raised about $369 million in its March 2026 Nasdaq IPO, giving it money for late-stage trials. It also has collaborations with Amgen and Novartis. It scores low on access because outsiders cannot license its platform.

Pick it if you are a pharma company looking for an AI protein-design partner with late-stage proof. Skip it if you want software you can run yourself.

Isomorphic Labs, spun out of Google DeepMind and led by Demis Hassabis, has the strongest modelling pedigree in the field. It co-developed AlphaFold 3, which predicts how proteins interact with drugs, DNA and antibodies. In February 2026 it unveiled its Drug Design Engine (IsoDDE), which it says more than doubles AlphaFold 3's accuracy on the hardest protein-drug predictions and matches physics-based binding methods at far lower cost. Those are company benchmarks, not independent tests.

Money and partners are not a problem. It raised $600 million in 2025 and a $2.1 billion Series B in 2026, and has research deals with Eli Lilly, Novartis and Johnson & Johnson. It appointed a chief medical officer in June 2025.

The gap is clinical proof. We found no disclosed drug candidate in human trials, so its top-tier technology scores are held back by the heaviest-weighted criterion.

Pick it if you are a large pharma partner wanting frontier structure-based design. Skip it if you need proof that the platform has already produced a drug that works in people.

Chai Discovery is the fastest-rising AI model builder in drug discovery. Its first model, Chai-1, predicts the 3D structure of proteins and their partners and was released publicly in September 2024. Chai-2, launched in June 2025, designs antibodies from scratch: the company reports double-digit success rates in lab tests, compared with the tiny hit rates of traditional screening. That is a vendor claim, but it helps explain an unusual run of 2026 deals with Eli Lilly, Pfizer, Novartis, argenx and Bristol Myers Squibb.

Funding has followed: a $130 million Series B at a $1.3 billion valuation in December 2025 and a $400 million Series C in July 2026.

Like Isomorphic, it has no clinical proof yet. Its business model is to supply design capability to pharma rather than run its own trials, so clinical validation will come through partners and may take years to show.

Pick it if you are a biologics team that wants AI-designed antibody starting points. Skip it if you need evidence of drugs already in patients.

Iambic, based in San Diego, combines two model families. NeuralPLexer predicts the 3D shape of protein-drug complexes, and Enchant, now at version 3, is trained to predict how molecules will behave in the body and in trials, not just whether they bind. The aim is to kill weak candidates earlier and cheaper.

Its lead drug, IAM1363, targets HER2, a well-known cancer driver, and is designed to be selective and to reach the brain, where HER2-positive breast cancer often spreads. It entered a Phase 1/1b trial in March 2024 and early data were presented at ESMO in October 2025. It is still an early-stage trial of a drug for a validated target.

Partners are arriving quickly: Takeda, Bayer and AbbVie all signed collaborations in 2026. Funding details are less public than for bigger rivals.

Pick it if you want a small-molecule partner with a clinical asset and outcome-prediction models. Skip it if you need later-stage proof or software you can license.

Xaira launched in April 2024 with $1 billion, one of the largest starting sums in biotech history, and is led by Marc Tessier-Lavigne, a neuroscientist and former Stanford president. It builds predictive and agentic AI models across discovery and development, and generates its own large biology datasets through an effort it calls X-Atlas. It won a 2026 Fierce AI Innovation Award for preclinical development.

On our criteria it scores lowest because there is little public to judge. We found no disclosed clinical candidate, no major pharma partnership and no licensable platform. That is normal for a company only two and a half years old, and its funding means it can take time.

We include it because buyers and investors often ask about it, and because its combination of money, talent and data could move it up quickly once programmes are disclosed.

Pick it if you are tracking where the next wave of AI biotech will come from. Skip it if you need a partner or platform you can use today.