
Osmo says AI smell model reached human-level olfaction using 10 million labels
Osmo says its AI smell model reached human-level olfactory intelligence using a proprietary dataset of 10 million labels and 800 million scent tokens. The company reports commercial fragrance use and a win over in-house perfumers at a major consumer goods company, but the article does not describe independent validation. The claim matters because it tests whether smell can be modeled as a data language for product design and possible detection tasks.
Osmo, spun out of Google Brain by Alex Wiltschko in 2023, said its model detects and discriminates smells about as well as an average person and creates scents about as well as a junior perfume designer. The company said the system has already been used in the fragrance industry and, last month, outperformed in-house perfumers at one of the world's largest consumer goods companies. The company did not publicly detail the evaluation protocol or identify the customer.
The claim rests on data collection. Osmo said existing databases contained fewer than 200,000 human smell descriptions, so it built its own corpus using about 30 experienced sniffers near Newark Airport and around 70 trained labelers in the US and overseas. They use spectrometers and chemical sensors to map molecular compositions, producing more than 10 million human labels and 800 million scent tokens that encode molecules and concentrations. A bus-sized robot mixes and prints samples for research and customer work.
Osmo frames scent tokens like words in a language: changing molecule order or proportion changes the perceived odor, similar to how word order changes sentence meaning. That approach mirrors foundation-model methods used in text and images, but applied to a sensory domain where labeled data are scarce and chemical sensors are costly. The article does not describe independent benchmarking, so the central performance claim remains company-reported.
The technology matters first to fragrance and consumer products, where scent design is costly and tied to human expertise. Osmo said brands are using it to design and manufacture new scents, and said proposed future uses include airport threat detection and disease diagnosis. Those later uses would require validated deployment records, but the immediate effect is a data layer that can turn smell from a craft into a measurable, generative system.
Fragrance and consumer product teams can use it to generate and test scent profiles, while possible detection and diagnostic uses would depend on independent validation.
Sources (1)
- [1]This smell startup says its AI nose is as good as a human's(https://www.technologyreview.com/2026/11/18/1148816/smell-startup-osmo-ai-nose-good-as-humans/)