GLOSSARY
Frontier model
A frontier model belongs to the most capable group under a stated assessment framework; the designation depends on time, tasks and available comparison systems.
Capability depends on the task
A capable model may help investigate contradictory information or several possible responses during difficult negotiation preparation. For straightforward classification of routine enquiries, the same model class may add unnecessary effort. “Frontier” is therefore neither a permanent product property nor a guarantee for every use case. A meaningful comparison specifies the date, tasks and conditions. It also separates model capability from the quality of the information and tools supplied by the application, which influence the outcome too.
Additional capability must justify its cost
A model router could assign tasks to models according to their requirements. Its value also depends on recognising difficult cases reliably. Use the same evaluation to compare candidates. Consider critical mistakes, response time and cost including human checking. A more expensive model may make economic sense if it demonstrably avoids consequential errors; a less expensive one may be sufficient for clearly bounded work. Keeping the decision tied to observed task performance allows new models to be assessed without treating brands or leaderboard positions as enduring guarantees.