An AI model stops learning when training ends. Whatever it knew on its last day of training is what it knows for the rest of its deployment. That is one reason companies spend heavily on AI and still struggle to get results. General-purpose models are built to do everything reasonably well, not to excel at one company’s specific tasks, so a business ends up choosing between paying more for the most capable model and accepting weaker results from a cheaper one. Two Stanford computer science professors have started an AI lab to break that trade-off, and it has raised $110 million.

Noeri builds self-improving AI that gets better by working inside a company. Instead of training a model once and deploying it, the company pitches AI that accumulates experience from a customer’s own data and work and grows more capable the longer it is used. Noeri was founded in 2026 and is based in Palo Alto, with an office in Zurich.
Its AI first studies a company’s data and workflows. It then improves the whole system around a task rather than a single model: the instructions and skills given to the model, the tools it uses, which model handles each step and, where it pays off, the model’s parameters. Every change is explained and sent to a person for approval, much like a code review.
The system works with closed and open-weight models alike and adapts as new ones arrive. Customers keep ownership of their data and of the workflows and models they develop, subject to applicable third-party model licenses.
The company points to three differences. Its AI keeps improving from on-the-job experience after deployment, it optimizes instructions, tools and model choice for each task instead of only swapping models, and the customer owns what is built. Noeri says it is aiming for an ecosystem of many smaller AIs, each shaped to an organization’s problems, rather than a single superintelligence made and owned by one company. The name Noeri means “intelligence, in the plural.”
The same approach extends to science. Noeri’s AI forms hypotheses, runs experiments and learns from the results, working with scientists and engineers on problems from new molecules and proteins to complex technical systems. Open-ended search uses far more compute than a single answer, so the company argues that smaller, specialized models are what make sustained discovery affordable.
The approach is rooted in its founders’ research. Their joint work on test-time training showed that learning after deployment can open up new capabilities. Guestrin’s self-distillation work (SDPO) lets a model act as its own teacher, turning mistakes into lessons, while Choi’s work on agentic orchestration found that a small model directing other models and tools can beat frontier models at a fraction of the cost.
Co-CEO Yejin Choi is a MacArthur Fellow who was named to the TIME100 Most Influential People in AI in 2023 and 2025. A Stanford professor, she was previously a professor at the University of Washington and a senior director at the Allen Institute for AI, and her research focuses on making smaller AI models more capable. She is also an AI2050 Senior Fellow at Schmidt Sciences, where she works on pluralistic alignment.
Co-CEO Carlos Guestrin co-founded Turi, the machine learning platform Apple acquired in 2016, then built and led Apple’s central machine learning and AI organization. He also co-created XGBoost, one of the most widely used machine learning technologies. Choi and Guestrin were colleagues at the University of Washington before both moved to Stanford.
Noeri announced the $110 million round, led by Menlo Ventures. Madrona, Kindred Ventures, Wing Venture Capital and NEA participated, along with strategic investors Cisco Investments, NVIDIA and Snowflake. The round’s stage and valuation were not disclosed.
Guestrin said every company is on the same curve of paying more for the best model or settling for a cheaper one, and that specialization can move the whole curve. Menlo Ventures partner Venky Ganesan said most new AI labs ask the market to wait years for a breakthrough, while the two founders are doing world-class research and putting it to work inside companies at the same time.
Noeri is working with strategic partners across industries, including some of its investors, and says it chose a varied group so its technology learns to generalize before it specializes for individual customers. It is hiring researchers and engineers in Palo Alto and Zurich.
Where Noeri Sits Among Enterprise AI Players
In tailoring models to company data, Applied Compute retrains open-weight models on enterprise data and runs them for customers. It raised $350 million at a $3.25 billion valuation and passed a $100 million run rate in under 18 months.
The frontier labs are competing on large models themselves. Thinking Machines Lab has released its 594-billion-parameter Inkling 2 Small and is reportedly seeking a $5 billion round, and Reflection AI, founded by Google DeepMind alumni, has launched its first open-weight model, Beam. Noeri pitches something different: small, specialized AI that the customer owns, and a system that improves it automatically.
For more on competitors, see here.
MORE FROM THE POST
- Applied Compute Raises $350M at $3.25B as Run Rate Hits $100M in Under 18 Months
- Defensive Cyber AI Lab Corma Raises $125M Series A at $1B Valuation
- LangChain Raises $300M Series C at a $3B Valuation Led by CapitalG
- AI Evals and Post-Training Company Mercor Raises $750M Series D at $20B Valuation
- Ocient Raises $100M Series C to Run Governed Agentic AI on Petabyte-Scale Data
- AI
- Cisco Investments
- enterprise AI
- funding
- Kindred Ventures
- Madrona
- mega funding
- Menlo Ventures
- NEA
- Noeri
- Nvidia
- Overseas funding
- Snowflake
- Wing Venture Capital
Share
Most Read
- 1
- 2
- 3
- 4
- 5


Leave a Reply