WEDNESDAY · 18 NOV 2026 · 18:49 UTCRSSMY BRIEFINGS
MODELS18 NOV 2026 · 18:04 UTC · 6 MIN READ

Osmo Claims Its AI Model Matches Human Smell Ability

Osmo, a Google Brain spinout, claims its AI model achieves "human-level olfactory intelligence," detecting and discriminating smells like an average person and creating fragrances comparable to a junior perfumer. Founder Alex Wiltschko says the tech, trained on over ten million human scent labels, is already used by major fragrance firms and identified repellents stronger than DEET. Critics argue marketing claims lack documentation and the system struggles with contextual smell interpr

MODELS DESK
Osmo Claims Its AI Model Matches Human Smell Ability
Osmo, a Google Brain spinout, claims its AI model achieves "human-level olfactory intelligence," detecting and discriminating smel…
Osmo Claims Its AI Model Matches Human Smell Ability

Short answer: Osmo, a Google Brain spinout, claims its AI model achieves "human-level olfactory intelligence," detecting and discriminating smells like an average person and creating fragrances comparable to a junior perfumer. Founder Alex Wiltschko says the tech, trained on over ten million human scent labels, is already used by major fragrance firms and identified repellents stronger than DEET. Critics argue marketing claims lack documentation and the system struggles with contextual smell interpretation needed for medical diagnostics.

Osmo AI smell technology human level

A New Jersey lab tucked near Newark Airport has become the unlikely headquarters for teaching computers how to smell. Inside, a crew of about thirty seasoned sniffers works through scent after scent, jotting down descriptors like "nutty" or "marine" while spectrometers break those same aromas down into molecular data. The operation belongs to Osmo, a startup spun out of Google Brain in 2023 by founder Alex Wiltschko, who’s dead set on giving machines a sense of smell. He describes scent molecules as the planet’s original language-one most species use to communicate and that can unlock vivid human memories with a single whiff.

Lately, Wiltschko claims the company’s model has hit what he calls human-level olfactory intelligence. By his yardstick, that means the system detects and discriminates smells roughly as well as an average person and dreams up new fragrances on par with a junior perfumer. He says the tech is already in use across the fragrance industry and that last month it beat in-house perfumers at one of the world’s largest consumer-goods giants. Wiltschko pictures eventual rollouts for airport threat detection or disease diagnosis, though he admits those applications are still a ways off.

How Osmo solved olfactory data shortage

Cracking the model meant solving a brutal data shortage. Wiltschko says existing databases held fewer than two hundred thousand human smell descriptions. Osmo has since amassed more than ten million human labels and eight hundred million scent tokens, capturing the exact molecules and concentrations behind specific odors. Spectrometers and other chemical sensors run nonstop, analyzing everything from fruit and flowers to pantry staples and organs. Around seventy trained labelers in the U.S. and overseas spend their days sniffing samples and logging impressions using a standardized fragrance vocabulary.

The company also runs a smell-printing robot the size of a school bus that mixes ingredients to precise formulas, churning out samples for research and custom fragrances for paying clients. During a demo in Lisbon, Wiltschko typed a prompt asking for a scent like a Portuguese custard tart aimed at young men. The model spat out three candidate profiles with the necessary molecular recipes, and a finished sample would have been ready in about a week. Osmo says its tech has already created retail fragrances for heavyweights including Procter & Gamble.

Gates Foundation funds Osmo insect repellent

The Gates Foundation kicked in five million dollars at Osmo’s launch and later awarded a three-and-a-half-million-dollar grant to evaluate molecules that repel disease-carrying insects. The company identified eight substances more repellent than DEET. It’s also teamed up with StockX, the online clothing marketplace, on using smell to spot counterfeit sneakers by detecting chemical differences in shoe materials.

Experts debate Osmo AI smell claims

Not everyone’s buying the hype. Avery Gilbert, a smell scientist and author, has taken Osmo to task on his Substack, arguing that some marketing claims lack documentation and that the company’s patent filings still lean on human sensory panels and perfumers rather than pure AI. A rep for Osmo didn’t respond to a request for comment before publication.

Michael Schmuker, a former professor of neural computation at the University of Hertfordshire who’s consulted for Osmo, calls the technology a genuine achievement. With colleagues, he built an open-source, reverse-engineered version of Osmo’s 2023 model, though it lacks the massive dataset the company has collected since. Still, Schmuker warns that human olfaction leans heavily on context. The same molecule can trigger completely different associations depending on the setting. He points out that a compound found in Camembert cheese also shows up in used nappies, yet people never confuse the two because context tells them which is which.

Inside a controlled fragrance lab, that missing context might not matter much. But it could turn into a real headache if Osmo’s system becomes the olfactory channel for broader AI world models, as Wiltschko suggests. Picture a machine sniffing out food spoilage on a production line or detecting tumors the way dogs have in pilot studies. For that to happen, Schmuker says sensor tech will need to improve dramatically and the models will have to keep advancing. Medical diagnostics remains the biggest prize.

Wiltschko knows the history of companies that aimed straight for that peak and flamed out because the challenge is expensive and brutally difficult. His strategy is to build a foundation at the foothills, sending expeditions up each mountain by essentially smelling the whole world. Whether that approach ultimately reaches the summit is still an open question.

Frequently asked questions

What is Osmo and who founded it?

Osmo is a startup spun out of Google Brain in 2023 by founder Alex Wiltschko. Based in a New Jersey lab near Newark Airport, the company employs about thirty trained sniffers and uses spectrometers to build AI models that can detect, discriminate, and generate scents.

Has Osmo's AI actually achieved human-level smell ability?

Wiltschko claims the model matches an average person's ability to detect and discriminate smells and can create fragrances on par with a junior perfumer. The system reportedly beat in-house perfumers at a major consumer-goods company, though smell scientist Avery Gilbert argues marketing claims lack documentation and patents still rely on human panels.

How much smell data has Osmo collected?

Osmo has amassed more than ten million human smell labels and eight hundred million scent tokens, far exceeding the fewer than two hundred thousand descriptions previously available in public databases. Around seventy trained labelers in the U.S. and overseas sniff samples daily using a standardized fragrance vocabulary.

What commercial applications is Osmo already pursuing?

Osmo's technology has created retail fragrances for Procter & Gamble. The Gates Foundation awarded a $3.5 million grant after Osmo identified eight molecules more repellent to disease-carrying insects than DEET. The company also partnered with StockX to detect counterfeit sneakers by analyzing chemical differences in materials.

What are the main limitations of Osmo's current approach?

Smell perception heavily depends on context-the same molecule can smell like Camembert cheese or used nappies depending on setting. Michael Schmuker warns this could be problematic if Osmo's system becomes the olfactory channel for broader AI models. Sensor technology also needs dramatic improvement for medical diagnostics like tumor detection.

Source: MIT Technology Review

Reson8 Releases Rumble-2-Flash 0.5B Speech Model for 40 Languages
models

Reson8 Releases Rumble-2-Flash 0.5B Speech Model for 40 Languages

Reson8 released Rumble-2-Flash, a 0.5-billion-parameter speech recognition model covering 40+ languages. Launched November 18, 2026 on Hugging Face under Apache 2.0, it runs in standard Transformers and vLLM without custom code. Distilled from the larger Rumble-2 via the GOLD on-policy method, it scores 6.96% error on the multilingual Open ASR Leaderboard, beating Meta's 7B omniASR-LLM and Whisper large-v3, and processes an hour of audio in about half a second on a single GPU.

Google Launches Free Tasks AI Agent Replacing Spark
models

Google Launches Free Tasks AI Agent Replacing Spark

Google has rebranded its Spark AI agent as Tasks and made it free for users in the US, Canada, Japan, Korea, Australia, and India. Powered by Gemini 4 Hydrogen, the agent handles planning chores like meal plans and calendar organization. US users gain additional capabilities, including calling businesses, making purchases via Google Pay, and delegating tasks via email or SMS. Each session runs in an isolated virtual machine, pausing for confirmation before sensitive actions.

Microsoft Moves Office Apps Into Copilot As Autonomous Agents Arrive
tools

Microsoft Moves Office Apps Into Copilot As Autonomous Agents Arrive

Microsoft integrated Word, Excel, and PowerPoint into Copilot, making the AI assistant the main work interface. Autonomous Autopilot agents handle tasks across email and calendars using Work IQ's context. Collaboration moves to Copilot Workspaces for shared human-agent goals. Governance relies on unique agent IDs, admin controls, activity logs, and the Agent 365 control plane. Usage is metered via Copilot Credits. An AI Transparency Initiative with partners aims to set safety standards

Share this post
Report