Models are out, molecules are in as AI drug startups sprint to the clinic

Across leading AI drug discovery companies, more than 30 AI-derived drugs were in human clinical trials by April 2024, with at least nine in Phase II or beyond, signaling a shift from model-building to molecule-driven value creation.6

Insilico Medicine’s INS018_055 (also reported as ISM001-055), discovered and designed by its Pharma.AI platform, became the first fully AI-designed molecule to reach Phase 2 trials for idiopathic pulmonary fibrosis, with Phase IIa data showing significant lung-function improvements.2 3 6

Absci advanced ABS-101, an AI-designed anti-TL1A antibody for inflammatory bowel disease, into a randomized Phase 1 trial in May 2025, marking its transition to a clinical-stage AI biologics company.1

Recursion Pharmaceuticals and other early AI-first players have moved multiple AI-discovered candidates into the clinic, but mixed efficacy signals (for example in neurovascular indications) fuel investor and industry skepticism about whether current AI approaches can deliver truly differentiated drugs.4 6

Newer startups are emphasizing direct translation to small molecules and oral drugs rather than only predictive models or biologics, exemplified by PsiThera, which uses AI to turn successful injectable TNF biologics into orally available small-molecule immune therapies.4

Industry commentary in 2024–2025 frames a strategic pivot:
investor enthusiasm is now tied less to abstract AI platforms or large models and more to tangible clinical assets—molecules in trials with human data and clear regulatory paths.1 2 6

BiopharmaTrend and other trackers highlight companies like Absci, BioAge, Generate:
Biomedicines, Insilico, Schrödinger, and Moderna as leaders because they pair AI platforms with advancing internal pipelines, including multiple AI-enabled antibodies, small molecules, and mRNA vaccines in Phase 1–3.1

Funding and partnerships increasingly reward AI startups that can both generate molecules and de-risk them via biology (e.g., automated labs, high-throughput experiments, and trial-outcome prediction) rather than offering software-only model services.1 2 6 7

Skeptics note that, despite faster discovery timelines and lower preclinical costs claimed by AI companies, there are still relatively few clear-cut efficacy wins in mid- to late-stage trials, so 2025–2027 is widely seen as the period when the “models vs. molecules” debate will be resolved by clinical data.4 6

Overall, current news and analysis portray a maturing AI-biotech sector where the key differentiator is no longer who has the largest or most sophisticated model, but who can consistently turn AI output into clinically validated molecules that improve patient outcomes.1 2 4 6