The foundation model for transactional behavior.
At Unbox AI, we are building a new class of foundation models for sequential human behavior in retail, payments, and finance. Large Behavioral Models (LBMs) that (1) learn directly from transactional behavior, (2) generalize across users, merchants, and geographies, (3) transfer to downstream applications including personalization, credit scoring, and fraud detection. Our model, BehaviorGPT, is available to developers with a free API key. Ultimately, we're pursuing general behavioral intelligence: what we think of as the AGI of human behavior in the digital economy.
Research
2026-06-03
Scaling Laws for Behavioral Foundation Models over User Event Sequences
We establish compute-optimal training recipes for behavioral foundation models across roughly 600 experiments spanning four orders of magnitude of compute. Behavioral models start data-hungry at small scale before converging toward the Chinchilla heuristic, and the metric you optimize shapes the scaling law itself.
2026-02-05
Large Behavioral Models: A Foundation Model Paradigm for Human Actions
We introduce Large Behavioral Models (LBMs), foundation models trained on chronological sequences of human actions. BehaviorGPT, a frontier LBM, delivers double-digit sales uplift for retailers and payment companies.
2025-07-15
BehaviorGPT: aesthetic purchasing preferences from behavior
We learn aesthetic preferences from purchasing and engagement sequences (how people discover, save, and acquire art and design) rather than from static labels alone. Trained on 215 billion human interactions across major art and design platforms.
FAQ
What does Unbox AI do?
Unbox AI builds BehaviorGPT, a Large Behavioral Model: a foundation model trained on sequences of human actions such as purchases, searches, and payments. One pretrained model powers fraud detection, recommendations, search, credit scoring, and lifetime value prediction across retail, engagement, and payments.
What is a Large Behavioral Model?
A Large Behavioral Model (LBM) is a foundation model trained on chronological sequences of human actions instead of text. It predicts the next action from the actions before it, and what it learns transfers to tasks like fraud detection and recommendations. Unbox AI introduced the term, and BehaviorGPT is its LBM.
How is a Large Behavioral Model different from a Large Language Model?
A Large Language Model learns from text and predicts the next word. A Large Behavioral Model learns from event sequences and predicts the next action, deciding in a single step rather than token by token. For ranking, BehaviorGPT-v4 scores a whole catalogue in 0.7 ms, 48 times faster than a prompted LLM.
What results has BehaviorGPT shown?
BehaviorGPT-v4 is a 12.5B-parameter model pretrained on 150 billion user actions. Zero-shot, it beats every baseline trained on up to 2.7 million samples of the target data, and one checkpoint leads all 13 public datasets it was tested on. In production A/B tests, BehaviorGPT models have increased sales by double-digit percentages against incumbent systems.
How can I try BehaviorGPT?
Developers can get a free API key on the BehaviorGPT page at unboxai.com/behaviorgpt. The key arrives by email immediately, with a link to the live demo and example notebooks. Enterprise deployments are available on request.
Who founded Unbox AI and where is it based?
Unbox AI was founded in 2019 by Rickard Bruël Gabrielsson and Gunnar Carlsson. It is headquartered in Boston, with offices in Stockholm and San Francisco.
Where can I read Unbox AI's research?
Papers on Large Behavioral Models, scaling laws, and each BehaviorGPT generation are published at research.unboxai.com.