Osmo says its smell AI matches a human nose: who pays and who gains
Osmo, a startup training AI on smell, says its model now matches an average person's nose. Fragrance brands and investors are paying; medical uses are still a hope.
Osmo, a startup building AI for smell, says its model has reached "human-level olfactory intelligence". The claim is the company's own, and so far the paying customers are in fragrance, not medicine.
The company defines the term narrowly. It means the model is about as good as the average person at detecting and discriminating between smells, and about as good as a junior perfume designer at creating them. That's a modest bar, which I think is the honest way to frame it. Last month, the company says, the model outperformed in-house perfumers at one of the world's largest consumer goods companies. It didn't name the company.
Who pays for a digital nose
Investors are footing the biggest part of the bill. Osmo completed a $70M Series B, bringing total capital raised to $130 million. Two Sigma Ventures led the round, and Stripe co-founder Patrick Collison took part. The 2023 launch was backed by a $60m Series A led by Lux Capital and Google Ventures.
Philanthropy has paid for the non-fragrance work. The Gates Foundation invested $5 million at launch, then gave a $3.5 million grant to assess molecules that repel disease-carrying insects; Osmo found eight substances more repellent than DEET.
Customers are the third source. Osmo sells fragrance development to brands, and its site pitches an end-to-end service from formulation through manufacturing, filling and packaging. Its technology has already been used to create retail fragrances, including for Procter & Gamble. You can see the pitch on Osmo's own site.
Who gains if it works
Brands come first. Osmo's pitch is speed: shrinking fragrance product development from months to days. In a demo, a text prompt for a Portuguese custard tart scent aimed at young men produced three scent profiles in a few minutes, with delivery of a physical sample about a week later.
Osmo gains too, though maybe not in the way you'd expect. CEO Alex Wiltschko has said that "If we're successful, you will know our name, but you'll never see our product." That's an ingredient-supplier model, closer to a chip designer than a consumer brand. I'd guess it's a sensible way to make money from a technology that customers never see.
The moat is data. Osmo says it has built a dataset of more than 10 million human labels and 800 million "scent tokens", with around 70 trained labelers sniffing samples and writing down what they smell like. Wiltschko's argument is that before Osmo, existing databases held fewer than 200,000 human descriptions of smells. If that's right, rivals would likely need years to catch up.
The doubters
Not everyone's sold. Smell scientist Avery Gilbert has argued on his Substack that some of Osmo's marketing claims are undocumented, and that its patent filings still rely on human sensory panels and perfumers rather than AI. Osmo didn't respond to MIT Technology Review before publication.
Even a friendlier voice has reservations. Michael Schmuker, who has consulted for Osmo, says context matters in smell: the same molecule can evoke different associations depending on the setting. A fragrance lab is a controlled setting, so that gap probably won't hurt there. Airports and hospitals are another matter.
Sources also disagree on small details. MIT Technology Review says Wiltschko spun the company out of Google Brain in 2023, while funding announcements describe a 2022 founding. I'd treat that as a launch-versus-founding difference, not a dispute.
The lesson for AI startups
The big talking point is diagnostics and threat detection. Wiltschko has called scent-powered health-care diagnostics Osmo's "Mount Everest", for example a tool that could do the work of medical service dogs. He's frank about the odds: he says many companies have gone straight at that peak and failed because it's expensive and difficult. Schmuker adds that sensor technology will need to get a lot better.
So the real story is a familiar one: a foundation-model playbook (lots of data, a narrow first market) applied to a sense nobody had digitised at scale. Fragrance pays the bills now. Medicine is the pitch to investors.
Honestly, I'd judge Osmo on whether its fragrance results are independently documented, not on the human-level label. A junior perfumer's skill is a fair benchmark, but it's the company grading itself. If the consumer goods win gets named and measured, I'd take the diagnostics talk more seriously. Until then, it's a promising fragrance supplier with a very large ambition.
