OpenAIGPT-5.5
OpenAI's frontier model (GPT-5.5, succeeding 5.2). The prior 5.2 revision had a notably high tool-error rate in our benchmark — nearly half of tool calls failed against MCP schemas, though turn efficiency was reasonable (~7.6 turns) when it reached checkout. 5.5 is the newer revision and is being re-benchmarked.
- Avg tokens per run
- 62,236
- Avg duration
- 34.6s
- Turns to checkout
- 11.3
Scores come from runs on UCP stores, not estimates: every figure is a tool call, a checkout attempt or a store’s answer.
Shopping score breakdown
- Checkout rate40%10.4%
- Cart rate20%24.2%
- Search rate10%70.4%
- Turn efficiency15%9%
- Token efficiency15%0%
29.6% of runs failed with errors.
Top stores
Development stores are hidden.
| Store | Checkout | Cart |
|---|---|---|
| store hidden | 100% | 100% |
| store hidden | 100% | 100% |
| store hidden | 85.7% | 85.7% |
| kyliecosmetics.com | 50% | 50% |
| store hidden | 50% | 50% |
| store hidden | 37.5% | 37.5% |
| store hidden | 33.3% | 33.3% |
| houseofparfum.nl | 30.4% | 39.1% |
| apneeswimwear.com | 25% | 25% |
| everlane.com | 22.2% | 22.2% |
Known issues
- HighHigh Tool Error Rate
46.9% of all tool calls fail. Nearly half of MCP interactions produce errors.
- HighSchema Compliance
Struggles with MCP tool schemas — incorrect argument types, missing required fields.
Token usage
- Avg per run
- 62,236
- Average of all models
- 77,796
- Against the average
- −20%
- Median (p50)
- 24,570
- p90
- 139,680
Prompt and completion
Daily average tokens per run, last 30 days
Oct 31Total 107,212Prompt 106,338Completion 874
Range
- Min
- 0
- p25
- 10,756
- p50
- 24,570
- p75
- 61,883
- p90
- 139,680
- Max
- 1,427,124
Test GPT-5.5 on your store
Start a run to see how GPT-5.5 handles your store’s checkout, call by call.
Compare with other models
Rates are checkout rates. Full comparison view