Model selection / Explore the tradeoffs
Find the shape of your pool.
Latency, task cost, and every benchmark that matters to the invocation.
Illustrative dataFictional models and scores demonstrate the selection geometry. No inference is dispatched.
6models in view
5on the frontier
1dominated
0outside bands / unknown
A projection of 5 dimensions
The plot shows two axes. Frontier decisions use all enabled benchmarks, cost, and latency.
0 models omitted for unknown coordinates. Dashed lines show band boundaries. Use the table for overlapping points.
FrontierDominatedOutside bands / unknown
Every axis, side by side
| Model | Pool status | |||||
|---|---|---|---|---|---|---|
| Illustrative endpoint | $0.0003 | 1.0s | 62 | 65 | 57 | Frontier |
| Illustrative endpoint | $0.0007 | 0.7s | 70 | 60 | 62 | Frontier |
| Illustrative endpoint | $0.0030 | 1.5s | 85 | 83 | 80 | Frontier |
| Illustrative endpoint | $0.0040 | 2.0s | 75 | 70 | 70 | Dominated |
| Illustrative endpoint | $0.0060 | 2.1s | 82 | 75 | 94 | Frontier |
| Illustrative endpoint | $0.0090 | 4.2s | 89 | 96 | 69 | Frontier |
Selected model
Balanced
Synthetic values for interaction design. Not real model evaluations.
Why it remains in the pool
Meets every band. No other eligible model is better or equal on all enabled axes and strictly better on one. A stronger benchmark may trade off against cost or latency.
This is pool exploration, not an award. Weighted preferences, pool-size limits, current budget, capabilities, and provider bids still determine the final route.