Built On Science. Powered By AI
January has been building glucose prediction models since 2020 and publishing the results. Explore our benchmarks, methodology, peer-reviewed research, and scientific advisors.
Measured against a real endocrinology board exam
88%
Highest endocrinology benchmark score
Our Health Context Engine outperforms GPT, Gemini, and human clinicians on the endocrine medical board exam. Retrieval runs against a curated internal corpus of clinical guidelines and peer-reviewed literature rather than the open web.
0.8%
Hallucination rate
The engine produced hallucinations in 0.8% of cases and contradictions in 1.8%, significantly lower than general-purpose frontier models tested on the same clinical cases.
2h
Forward glucose prediction
CGM-based glucose predictions 2 hours into the future, and virtual continuous glucose curves for people who don't wear a CGM.
Measured behavior change in the January app
Protein intake
+6.9%
Carb intake
-3.5%
Fiber intake
+13.8%
Sugar intake
-16.7%
Participants who lost weight
74.4%
Time in Healthy Glucose Range for people with T2D *
49.7% => 57.4%
* Every 5% increase in Time in Range is clinically meaningful for people with type 2 diabetes.
Six years of models, in order
CGM-based glucose predictions 2 hours into the future
Accurate post-CGM predictions
Virtual continuous glucose curves
Millions of data points from thousands of patients inform a probabilistic model without sensors
Health Context Engine built on a medical-grade personal health knowledge graph
Integrated patient application and clinical decision support tools
Peer-reviewed papers
AI-supported CGM app improved glycemic control and weight; T2D time in range rose from 49.7% to 57.4%
Read the paperGlucose prediction and behavior change without continuous sensors
Read the paperDigital program outcomes for people living with type 2 diabetes
Read the paperEvaluating a flexible, AI-supported digital diabetes program
Read the paperMachine-learning blood glucose prediction from CGM data
Read the paperMethodology reviewed by our AI and Scientific Advisory Boards. Meet them