Sol Trialbench · METHOD

How this tool works.

1. Use identical task inputs and settings where supported.
2. Score each result from 0–10 with one rubric; record duration and actual billed cost.
3. Paste paired rows. Keep failures as scored rows rather than silently excluding them.

Example, not a hidden assumption

The sample has one GPT-6.1 win, one GPT-6 win and one tie. Illustrative user-scored data, not a model benchmark.

The tool processes supplied inputs, not a verified full game database or a live model benchmark. No purchases, game-account actions or messages are sent.