BREAKING Inductive Bio's Beacon models beat 370+ submissions to win the world's largest ADMET challenge $25M Series A led by Obvious Ventures - a16z & Lux back the round ~80% accuracy on novel chemotypes vs. ~30% for the old way $21M award to model drug toxicity and cut animal testing NYC 708 3rd Avenue - small team, large benchmark BREAKING Inductive Bio's Beacon models beat 370+ submissions to win the world's largest ADMET challenge $25M Series A led by Obvious Ventures - a16z & Lux back the round ~80% accuracy on novel chemotypes vs. ~30% for the old way $21M award to model drug toxicity and cut animal testing NYC 708 3rd Avenue - small team, large benchmark
Inductive Bio logo
Company File - AI / Drug Discovery

Inductive Bio

The New York lab teaching machines to predict how a drug behaves in the body - before anyone bothers to make the molecule.

Founded 2023 New York, NY ~37 people $29.3M raised

Pictured: a wordmark on navy. Behind it, a quiet argument that competitors should share their homework.

Dispatch / Who They Are Now

A chemist clicks "run." A million experiments happen.

It is a Tuesday in a Manhattan office, and somewhere a medicinal chemist is doing something that used to take a year. She sketches a molecule that does not yet exist, asks a piece of software whether it will dissolve, survive the liver, and stay out of the brain - and gets an answer before her coffee cools. The molecule is still imaginary. The verdict is not.

That software belongs to Inductive Bio, a company whose entire reason for existing is to move the moment of truth earlier. Drug discovery has always been a long walk in the dark. Inductive turned on a flashlight and pointed it at the part of the path where most teams trip: the grind of making a promising compound actually behave like a drug.

They are not promising a cure. They are promising fewer dead ends - which, in this industry, is the more honest and more valuable thing to sell.

Inductive Bio sells a flashlight, not a miracle. In drug discovery, the flashlight is worth more.
The Problem They Saw

Whack-a-mole, with billions on the line

Roughly half of preclinical drug discovery is spent on a maddening balancing act. A molecule is potent but won't absorb. You fix absorption, and now it's toxic. You fix toxicity, and the potency drops. Co-founder Josh Haimson heard the same complaint from chemist after chemist, and it stuck.

"When they fix one issue with a molecule, two other issues pop up."- Josh Haimson, Co-founder & CEO

The properties at the center of this game have a clinical name: ADMET - absorption, distribution, metabolism, excretion, and toxicity. They decide whether a brilliant idea becomes a pill or a footnote. And historically, the only way to learn them was to make the molecule and test it in a lab. Slow. Expensive. Repeat a few thousand times.

The deeper problem was data. Every pharma company guarded its own experimental results like state secrets, so every model was trained on a sliver of reality. Smart people, starving datasets. Inductive looked at that arrangement and asked the slightly impolite question: what if the secrets were the bottleneck?

The Founders' Bet

Rivals should share their homework

Josh Haimson and Ben Birnbaum had built the machine learning organization at Flatiron Health, the oncology-data company Roche bought in 2018. They knew a particular trick well: the value isn't only in the algorithm, it's in the data nobody else has bothered to pool.

So they made a bet that sounds naive until you sit with it. Competing biopharma companies would contribute anonymized experimental data to a shared, pre-competitive consortium. In return, every member would get models trained on far more chemistry than any single company could ever generate alone. The real enemy, the bet goes, isn't the lab across town. It's the failed drug.

The competition isn't the lab across town. It's the molecule that fails in year nine.

It is an idea with a certain Wildean charm: the most selfish thing these companies can do, it turns out, is cooperate. Inductive Bio emerged from stealth in 2023 with $4.3M in seed funding co-led by a16z Bio + Health and Lux Capital, and started building the thing that would make the cooperation pay off.

The Short, Loud History

From stealth to state-of-the-art

2023

Out of stealth

Founded by Josh Haimson and Ben Birnbaum, ex-Flatiron Health. $4.3M seed co-led by a16z Bio + Health and Lux Capital.

2025

First competition win

Takes the Polaris ADMET challenge - the first public proof the approach beats incumbents on real benchmarks.

May 2025

$25M Series A

Led by Obvious Ventures, with a16z, Lux, S32, Character and Amino Collective. Total raised reaches $29.3M.

2025

ADME-One & a $21M toxicity award

Launches ADME-One with Ginkgo Datapoints and Tangible Scientific; lands an award of up to $21M to model drug toxicity and reduce animal testing.

Feb 2026

Wins the world's largest ADMET challenge

Beacon models place first among 370+ submissions in the OpenADMET-ExpansionRx blind challenge, across nine real-world endpoints.

The Product

A virtual lab with three personalities

Inductive's platform is less a single app than a small staff that happens to be made of math. The pieces are named, which tells you how the team thinks about them - not as features, but as colleagues.

Models

Beacon-1

The brain. ADMET-prediction models trained on the pre-competitive consortium data, tuned to hold up on novel chemotypes - the molecules nobody has seen before.

Software

Compass

The cockpit. Chemists run virtual experiments, review predictions, and decide what to synthesize in real time, without leaving the screen.

Assistant

Indy

The colleague. An AI chemistry assistant that automates the tedious operations and helps design and interpret experiments.

Platform

ADME-One

The bridge. A Tier-1 ADME plus human-PK projection offering, built with Ginkgo Datapoints and Tangible Scientific, pulling pharmacokinetics earlier into discovery.

Most software has features. Inductive's has names - because the team treats its models like staff, not switches.
The Proof

Numbers that survive a blind test

It is easy to claim accuracy on the molecules you trained on. The hard part - the only part that matters - is predicting compounds the model has never seen. That is exactly what a blind challenge measures, and Inductive keeps winning them.

Accuracy on novel chemotypes

Inductive's deep-learning approach vs. traditional methods // approximate
Inductive Bio (Beacon)0%
Traditional approaches0%
Source: Inductive Bio (figures approximate, self-reported on novel chemotypes).
370+
Submissions beaten in the OpenADMET blind challenge
#1
Finish, twice (Polaris + OpenADMET)
9
Real-world ADMET endpoints predicted
$29.3M
Total funding raised to date

Behind the benchmarks sit real users: biotechs like Aleksia, Architect Therapeutics, Arrakis, Belharra, Nested, Nexo, Rapport and Tenvie, working on everything from cancer to brain disorders. And behind them sit investors who tend to be allergic to hype - Obvious Ventures led the Series A, with a16z, Lux, S32, Character and Amino Collective along for the ride.

"These results affirm the value of our small but mighty team's experience in drug discovery and focus on building a diverse, high-quality dataset."- Josh Haimson, after the OpenADMET win
The Mission

Make the dead ends cheaper to find

Inductive's stated mission is to pair scientists with a virtual lab that runs millions of in-silico experiments, surfacing the strongest hypotheses worth a real test tube. Strip away the language and the goal is simple: spend less money learning that a molecule won't work.

That has a quieter consequence the company is now chasing directly. With an award of up to $21M, Inductive is building next-generation toxicity models - the kind that could reduce how often new drugs need to be tested on animals. Better predictions aren't only faster. Sometimes they're kinder.

Better predictions aren't only faster. Sometimes they mean one fewer animal in the experiment.
Why It Matters Tomorrow

Back to that Tuesday

Return to the chemist and her cooling coffee. The old version of her job was a series of expensive guesses, each one a synthesis, an assay, a few weeks of waiting to learn the molecule she loved had a fatal flaw. She made things in order to find out they were wrong.

The new version asks first. She tests a hundred ideas before lunch, kills the doomed ones on a screen, and walks into the lab to make only the molecules that earned the trip. The consortium she'll never meet has already taught her model what failure looks like.

That is the whole bet, sitting in one room: shared data, predicted early, with a human still holding the pen. Inductive Bio didn't make drug discovery easy. It made the guessing cheaper - and in a business where most molecules fail, cheaper guessing is how more of them eventually succeed.

Spread the word