# Osmo claims its smell foundation model matches human-level olfaction

> Osmo's smell AI matches the average person at detecting odors and a junior perfumer at creating them, trained on over 10 million human labels and 800 million scent tokens.

- Type: AI news
- Page: https://completeaitraining.com/news/osmo-claims-its-smell-foundation-model-matches-human-level/
- Published: 2026-11-19
- Updated: 2026-11-19
- Author: Jeroen Erne
- Jobs: Creatives, Healthcare, Marketing, Science and research, Writers
- Reading time: 4 min

AI startup Osmo says its smell foundation model has reached "human-level olfactory intelligence," matching the average person at detecting and discriminating odors and a junior perfume designer at creating them. The company, spun out of Google Brain in 2023, built its training data by having roughly 30 professional sniffers annotate a vast range of odors cross-referenced with chemical analysis, aiming to give computers a sense of smell.

In a lab near Newark Airport in New Jersey, a team of experienced human sniffers is working through thousands of odors. Their notes-"nutty," "peachy," "marine"-are paired with chemical analyses to create training data. Founder Alex Wiltschko's goal is straightforward: "to give computers a sense of smell."

"Scent molecules are really the only language most species on this planet can speak," Wiltschko said. "Most species can't communicate with sound or visually, so it truly is Earth's original language. And it's incredibly powerful for us, from a memory and an emotional standpoint. You can smell something once and remember it for the rest of your life."

The model is already being used in the fragrance industry. Last month, Osmo's AI outperformed in-house perfumers at one of the world's largest consumer goods companies. Wiltschko believes the technology could eventually be deployed for threat detection at airports or diagnosing diseases.

#### Building a dataset from scratch

Wiltschko ran into a familiar problem when applying the AI playbook to smell: there was almost no good data. Before Osmo, fewer than 200,000 human descriptions of smells existed in databases, by his count. Chemical sensor devices are expensive, and the fragrance industry guards its olfactory science closely.

The company's response was to build its own dataset-more than 10 million human labels and 800 million "scent tokens" that describe exact molecules and concentrations. Spectrometers and chemical sensors run around the clock, breaking down the molecular makeup of fruit, flowers, pantry staples, and organs. A team of about 70 trained data labelers in the US and overseas spend their days sniffing and recording observations using a standardized fragrance vocabulary.

Mapping scents this way reveals unexpected connections. Changing the order of words alters a sentence's meaning; rearranging scent molecules in different proportions does the same for smell. "The scent tokens that define the smell of a newborn baby have massive overlap with the scent tokens in a fresh plate of cooked spaghetti," Wiltschko said-specifically Barilla brand spaghetti.

#### From prompts to printed fragrances

Osmo creates its own scents using a smell-printing robot the size of a school bus. The machine mixes ingredients according to precise formulas to make samples for research and fragrances for paying customers. It can design and print scents to order based on a ChatGPT-style prompt.

Wiltschko demonstrated the system by generating a fragrance that smelled like a Portuguese custard tart, aimed at 18- to 23-year-old men. The system spent a few minutes processing, then offered three scent profiles-creamy, coconutty, fruity-along with the molecules needed to create them. A finished sample would be ready about a week later.

The technology has already produced retail fragrances for companies including Procter & Gamble. The Gates Foundation, which invested $5 million when Osmo launched, later gave the company a $3.5 million grant to assess insect-repelling molecules. Osmo identified eight substances more repellent than DEET. The company has also worked with StockX on using smell to spot counterfeit sneakers by detecting differences in the chemical composition of materials.

#### Skepticism and the problem of context

Not everyone is convinced. Avery Gilbert, a smell scientist and author of *What the Nose Knows*, has criticized Osmo on his Substack, arguing that some marketing claims are undocumented and that patent filings still rely on human sensory panels and perfumers rather than AI. "This heavy reliance on human sensory judgments seems odd for a company premised on super-cool AI," he wrote in March.

Michael Schmuker, a former professor of neural computation at the University of Hertfordshire who has consulted for Osmo, says the technology is an achievement. He and colleagues created an open-source, reverse-engineered version of Osmo's 2023 model, but it lacks the data the company has since collected.

Schmuker also points to a deeper challenge. "Smell is super complex. Context is very important for smell. The same molecule may evoke a different association depending on context," he said. The smell of Camembert cheese shares molecules with used diapers. People don't confuse the two because context separates them. In a controlled fragrance lab, this may not matter. It could become a problem if Osmo's technology is deployed as the "olfactive channel" for AI world models-machines that sniff out food spoilage on production lines or detect tumors through smell.

For medical diagnostics to work, sensor technology must improve substantially. That remains the biggest prize. "There's a lot of history of companies that have gone directly at that mountain peak and, unfortunately, not been successful, because it is very expensive and it is very difficult," Wiltschko said. "So what we're doing is building this foundation at the foothills of all these mountains."

#### Why this matters for creatives, marketers, and researchers

Osmo's work signals that scent is becoming a programmable medium, not just an artisanal craft. For fragrance marketers and product designers, AI-generated scent profiles could compress the development cycle from months to days. Researchers in healthcare and food science should watch how the underlying approach-pairing human sensory labels with chemical analysis at scale-could translate to taste, texture, or diagnostic applications. Writers and creatives working on multisensory projects may find new vocabulary and conceptual models as the "scent token" approach makes smell as describable and composable as words in a sentence. The [AI for Scientists Courses](https://completeaitraining.com/tag/science-and-research/) cover the kind of data-labeling and model-training techniques that underpin projects like this one.

---

This is the Markdown form of https://completeaitraining.com/news/osmo-claims-its-smell-foundation-model-matches-human-level/. The whole site is mapped at https://completeaitraining.com/llms.txt.
