Stop ranking models. Map the tradeoff frontier.
Use comparable measurements from one workload. The bundled names, prices, and scores are fictional and make no current provider or access claim.
Budget per workload
Prediction at $8
Fictional five-model sample ready.
Use the same workload, score definition, and cost basis for every row. Mixed benchmarks make the frontier meaningless.
Frontier points4 / 5
Dominated1
Best within budgetOpen B
Selected quality85.0
BEST UNDER $8.00: Open B · quality 85.0Prediction confirmed.
Legacy D is dominated by Open B: $6.00 cheaper and 7.0 quality points higher.
FRONTIER · Tiny E → Open B → Closed A → Closed C
Why: raising the budget can change the feasible winner, but it never makes a dominated point efficient.