What is a frontier model?
The Harvard Chan AI Glossary
In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.
A frontier model is an AI model near the leading edge of capability at a particular time. A model is a trained system that produces predictions or generated responses from supplied information. “Frontier” is a comparative description: it concerns what is currently achievable, not membership in a permanent technical category.
The boundary moves. A model considered exceptionally capable one year may later become ordinary. Different tests also emphasize different strengths. A model that performs particularly well on difficult mathematics may not be the best choice for rapid transcription or a small device with limited memory.
Capability needs to be specified. Does the model write code effectively, interpret images, handle unfamiliar questions or work through long tasks? A benchmark is a standardized set of tasks used to compare systems. Benchmarks can provide useful evidence, but their results do not automatically establish performance on a particular organization’s work.
The surrounding application matters too. A model given a calculator, relevant documents and clear instructions may perform differently from the same model used alone. Comparing complete products without identifying those differences can make it difficult to tell what the underlying model contributed.
Do not choose the newest model by default. Choose one that performs the task well enough at an acceptable cost. Here, trust should mean demonstrated performance under understood conditions, not familiarity with a brand. A smaller or less expensive model may be sufficient when the job is narrow and the required answer is straightforward.
There can be a real trade-off between a more capable model and the time or cost required to use it. More importantly, a stronger model cannot supply information that the application never gives it or fix a broken connection to the system it must use.
A health system comparing models for prior-authorization packet preparation should test the same clinical notes, payer criteria and required outputs rather than simply select the newest frontier model. A more capable model may help synthesize complex guidelines, but it cannot recover a lab result the application never supplies. Compare errors, correction effort, response time and total cost. “Frontier” means near the leading edge of specified capabilities; it is not a clinical quality certificate.
If you would like to find out more about AI and improve your leadership skills, consider the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.
AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.
For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.
Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy