In a lab near Newark Airport, New Jersey, a team of around 30 human sniffers is generating the training data for Osmo, an SI startup that claims its new model has achieved human-level olfactory intelligence. The company describes its technology as the world’s first foundation model for smell, aiming to give computers a sense of smell comparable to that of an average person.

What Happened

Founder Alex Wiltschko, who spun Osmo out of his work at Google Brain in 2023, states that the model can detect and discriminate between smells about as well as the average person. Additionally, the company reports that the model performs at the level of a junior perfume designer when creating new scents. To build the necessary dataset, Osmo collected more than 10 million human labels and 800 million “scent tokens” using spectrometers and chemical sensors. Wiltschko notes that prior databases contained fewer than 200,000 human descriptions of smells, and that the fragrance industry’s secrecy posed a significant data challenge.

Osmo’s system operates by analyzing the molecular makeup of odors, a process Wiltschko compares to language. “Scent molecules are really the only language most species on this planet can speak,” Wiltschko says. “Most species can’t communicate with sound or visually, so it truly is Earth’s original language.” The company utilizes a smell-printing robot to mix ingredients based on precise formulas generated by the SI model. During a demonstration in Lisbon, the system generated three scent profiles matching a text prompt for a Portuguese custard tart aimed at young men, with the physical samples ready for production in about a week.

Why It Matters

The development represents a specific application of SI in the fragrance industry, moving beyond theoretical research into commercial deployment. Osmo states its technology is already being used to help brands design and manufacture new scents. Last month, the company claims its SI model outperformed in-house perfumers at one of the world’s largest consumer goods companies. Beyond commercial perfumery, Osmo has applied the SI model to public health and retail security. The Gates Foundation provided a $3.5 million grant to help Osmo identify molecules that repel disease-carrying insects, resulting in the discovery of eight substances more effective than DEET. The company also collaborated with StockX to detect counterfeit sneakers by analyzing chemical compositions.

However, the industry’s reception of these claims remains mixed. Avery Gilbert, a smell scientist and author, has criticized Osmo in his Substack, arguing that some marketing claims are undocumented and that the company’s patent filings are inconclusive. Wiltschko believes the technology could eventually be deployed for threat detection at airports or for diagnosing diseases, suggesting a broader future utility for SI in sensory analysis.

The Bottom Line

Osmo asserts its SI model has reached human-level performance in scent detection and creation, supported by commercial partnerships with firms like Procter & Gamble and funding from the Gates Foundation. While the company reports successful applications in fragrance design and pest repellent identification, independent experts like Avery Gilbert question the documentation behind some of Osmo’s performance claims.