Osmo

Osmo is the most serious attempt yet to teach a computer to smell, and it is here because the science is real rather than because the pitch is loud. It span out of Google Brain in 2022 under Alex Wiltschko, an olfactory neuroscientist with a Harvard PhD in the subject, and its stated mission is to digitise the sense of smell.

The credential that matters is a 2023 paper in Science, co-led with the Monell Chemical Senses Center, introducing the Principal Odor Map: a graph neural network trained on roughly 5,000 molecules that learned to predict what a molecule smells like from its structure. On a prospective set of 400 odorants it had never seen, the model matched the trained panel's average description more closely than the median human panellist did. That is the whole field's high-water mark, and it is peer-reviewed rather than a press release.

The commercial arm, launched in 2025 and originally badged Generation by Osmo, is a fragrance house where the first draft comes from a model. You give it a prompt - text, an image, a piece of audio - and its Olfactory Intelligence returns a formulation, which in-house perfumers then refine. Christophe Laudamiel, whose credits include Tom Ford and Thierry Mugler, is among the noses on that side, which tells you the human step is not decorative.

Worth being straight about the shape of it: this is business to business. Osmo sells fragrance design to brands and creators, not bottles to you. There is no consumer price list and nothing to buy, and the standalone Generation site now redirects to the parent. Read it as infrastructure for the industry you buy from, not as a house you can shop.