Osmo’s AI Nose Reaches Human-Level Smell Detection, Paving Way for Fragrance and Security Innovations
Osmo’s AI nose now matches human ability to detect and create scents, opening new possibilities in fragrance design, security, and health diagnostics.
S
··2 min readSciTech Editorial Team
Newsroom

In a modest laboratory near Newark Airport, Osmo’s artificial‑intelligence (AI) platform announced that its scent‑recognition model now operates at “human‑level olfactory intelligence.” Founder Alex Wiltschko explained that the system can detect and discriminate odors as accurately as an average person and compose new fragrances with the skill of a junior perfumer, a milestone that could reshape industries ranging from cosmetics to security.
The breakthrough rests on an unprecedented dataset: more than 10 million human‑generated scent labels paired with 800 million “scent tokens” that encode exact molecular identities and concentrations. Osmo’s engineers employ spectrometers and other chemical sensors around the clock to decompose everything from fruit to industrial chemicals, while a global crew of roughly 70 trained annotators records nuanced descriptions—“nutty,” “peachy,” “marine”—using a standardized fragrance lexicon. A robot the size of a school bus mixes ingredients on demand, allowing the AI to generate scent formulas from simple text prompts, much like a ChatGPT‑style interface.
Already, the technology is being commercialized. Procter & Gamble has used Osmo’s model to accelerate new perfume development, and the Gates Foundation awarded a $3.5 million grant that enabled the AI to identify eight compounds more repellent than DEET for disease‑carrying insects. StockX partnered with Osmo to detect counterfeit sneakers by spotting subtle chemical differences, while the company envisions future deployments at airports for threat detection and even medical diagnostics based on volatile biomarkers.
The claims have not gone unchallenged. Smell scientist Avery Gilbert criticized Osmo’s marketing as insufficiently documented, noting that its patents still rely on human sensory panels. Wiltschko retorted that “smell is ultimately a human experience; our model must learn from humans to bridge chemistry and perception,” arguing that reliance on human data is a feature, not a flaw. Former neural‑computation professor Michael Schmuker acknowledged the achievement but warned that open‑source replicas lack the massive data volume that gives Osmo its edge.
Looking ahead, Osmo plans to expand its sensor network, reduce the cost of data acquisition, and refine its model to handle complex, real‑world scenarios such as diagnosing respiratory diseases from breath. If it can sustain the current growth trajectory, the AI nose may soon become as ubiquitous as vision‑based systems, turning scent into a new frontier for digital intelligence.



