Best AI model for Model ladder: fix 10 real merged bugs with tests hidden
Claude Haiku 5.5 ranks first for “Model ladder: fix 10 real merged bugs with tests hidden” with a score of 92.6, passing 100% of its 5 runs.
Lowest average cost per success: Claude Haiku 5.5 at $0.0084.
Based on 15 logged runs; 3 models had enough runs to rank.
| # | Model | Score | Runs | Pass rate | Cost / success | Avg time | Step-ins / run |
|---|---|---|---|---|---|---|---|
| 1 | Claude Haiku 5.5 | 92.6 | 5 | 100% (5/5) | $0.0084 | 38s | 0.00 |
| 2 | Claude Sonnet 5.5 | 85.9 | 5 | 100% (5/5) | $0.07 | 57s | 0.00 |
| 3 | Claude Opus 5.5 | 76.5 | 5 | 100% (5/5) | $0.25 | 1m | 0.00 |
Only models with 3+ runs here are ranked; all settings of a model are grouped. Scores use the default weights (methodology).