Recursion's AI engine finally has clinical data to point to

The technology

Recursion uses automated cell imaging, large biological datasets and machine learning to find drug targets and design molecules. Fierce Biotech reported that it holds 50 petabytes of data and makes about 330 compounds for each one that reaches the clinic, against an industry figure of 2,500 to almost 5,000. Khan led the company's merger with Exscientia, Fierce reported, and the company now pitches itself as an end-to-end discovery and design shop.

The platform now has two clinical signals. REC-4881, a MEK1/2 inhibitor, cut total polyp burden in 87% of 15 FAP patients with a 47% median reduction after 12 weeks. REC-1245, an RBM39 degrader Recursion says it designed start to finish, produced three partial responses at 45 mg or higher, two confirmed in endometrial cancer. Partners are also taking assets: Genentech exercised its first neuroscience target option and Sanofi optioned its first development candidate.

The odds

Both lead assets are early. TUPELO is a small, open-label, single-arm phase 1b/2 study. The BIO dataset puts the chance of moving from phase 2 to phase 3 at 28.9%, and rare disease programs reach approval from phase 1 about 17.0% of the time. FAP is a rare disease, which helps, but the FDA has not yet said what endpoint a registrational trial should use. Recursion plans to start that trial in the second half of 2027, pending FDA talks.

REC-1245 sits in phase 1, where the overall chance of approval is 7.9%. Three responses in dose escalation is a start, not a trend. The earlier oncology bets, REC-617 and REC-3565, are now up for partnering.

The money

Recursion had about $561 million in preliminary cash at Sept. 30, 2026, then raised about $200 million by selling 61.54 million shares at $3.25. It says the money lasts into 2029. The second-quarter net loss was $131.0 million on $7.7 million of revenue, and 2026 cash operating expense guidance is under $375 million.

Partner income is real but uneven. Recursion has received $216 million from Roche and Genentech and $144 million from Sanofi to date. The Genentech deal covers up to 40 programs, each with more than $300 million in potential milestones.

The risks

The first risk is REC-4881's path to registration. A single-arm dataset of 15 efficacy patients leaves room for the FDA to ask for a larger or controlled trial, and four of 26 patients stopped treatment because of adverse events. The second is dilution: the October deal priced shares at $3.25, and the company will need more capital or large milestones before any product sale. The third is that revenue depends on partners' schedules, as the drop from $19.2 million to $7.7 million in quarterly revenue showed.

Our read

Recursion has moved from promising many medicines to showing a few data points, and that is progress. REC-4881 is now a credible rare disease asset with a durable effect after dosing stops, and REC-1245 gives the platform a second, independent signal. The company has cash into 2029, so the next two years will be judged on clinical execution, not survival. The test is whether the FDA agrees to a registrational design for REC-4881 this year and whether REC-1245 holds its activity in a defined endometrial cancer cohort. Until then, the AI platform's value rests on two small datasets and a set of partner options that pay modestly up front. The October data and the cash raise make the next year more informative than any recent year for Recursion, because the questions are now about specific drugs, not about the promise of the platform.