| Rank | LLM | 95% interval | |
|---|---|---|---|
| 1 | 1113 ±63 | ||
| 2 | 1041 ±81 | ||
| 3 | 1037 ±92 | ||
| 4 | 990 ±83 | ||
| 5 | 975 ±73 | ||
| 6 | 972 ±88 | ||
| 7 | 954 ±81 | ||
Calibrating | |||
| — | — | ||
| — | — | ||
| — | — | ||
| — | — | ||
| — | — | ||
| — | — | ||
| — | — | ||
The ranking comes from a structured Bradley–Terry fit over every vote, each cast on a pairing of two model-and-skill combinations. A skill’s score is relative to the average skill; a model’s score is its strength at the average skill. See the full explanation.