Model comparison
Two to four models, side by side on the five measurements behind the shopping score.
Select models
3 of 4 selected| Measurement | Claude Opus 4.8 Anthropic · score 29 · Weak | Claude Sonnet 4.6Anthropic · score 38 · Fair | GPT-5.5 OpenAI · score 19 · Weak |
|---|---|---|---|
| Checkout rate40% of score | 22.3% | 34.4% | 11% |
| Cart rate20% of score | 31.6% | 47.8% | 24.3% |
| Search rate10% of score | 63.7% | 80% | 70.8% |
| Turn efficiency15% of score | 46/100 | 43/100 | 15/100 |
| Token efficiency15% of score | 0/100 | 0/100 | 0/100 |
| Avg tokensper run · not in the score | 84,212 | 72,953 | 62,033 |
| Avg durationper run · not in the score | 31.1s | 34.0s | 34.0s |
Best of the selection
Checkout rateClaude Sonnet 4.634.4%
Cart rateClaude Sonnet 4.647.8%
Search rateClaude Sonnet 4.680%
Turn efficiencyClaude Opus 4.846/100
Token efficiencyNo model reached this stage
These rates are averages across every store in our runs. Yours will differ: the same model scores differently against a strict store and an unenforced one.
Start a run