DeepSeekDeepSeek V3.2
The highest checkout rate in the benchmark at 68%. Efficient at 6.2 turns average with low token usage. The trade-off: a 13.1% tool error rate, suggesting it brute-forces through errors rather than avoiding them. Never needs nudging — pushes through autonomously. The best raw completion rate of any model tested.
- Avg tokens per run
- 75,971
- Avg duration
- 48.4s
- Turns to checkout
- 6.7
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%20.4%
- Cart rate20%36.8%
- Search rate10%75%
- Turn efficiency15%66%
- Token efficiency15%67%
24.3% of runs failed with errors.
Top stores
Development stores are hidden.
| Store | Checkout | Cart |
|---|---|---|
| store hidden | 77.8% | 77.8% |
| houseofparfum.nl | 72% | 72% |
| monos.com | 50% | 50% |
| ucp.travel | 33.3% | 33.3% |
| allbirds.com | 20% | 20% |
| snocks.com | 0% | 0% |
| store hidden | 0% | 50% |
| store hidden | 0% | 0% |
| store hidden | 0% | 100% |
| store hidden | 0% | 50% |
Known issues
- MediumTool Error Rate
13.1% error rate — brute-forces through errors rather than avoiding them.
- LowJSON Formatting
Occasional issues with tool argument types. Requires extra validation overlays.
Token usage
- Avg per run
- 75,971
- Average of all models
- 78,167
- Against the average
- −3%
- Median (p50)
- 30,870
- p90
- 147,357
Prompt and completion
Daily average tokens per run, last 30 days
Oct 21Total 1,620,285Prompt 1,614,646Completion 5,639
Range
- Min
- 2,517
- p25
- 18,429
- p50
- 30,870
- p75
- 58,517
- p90
- 147,357
- Max
- 1,620,285
Test DeepSeek V3.2 on your store
Start a run to see how DeepSeek V3.2 handles your store’s checkout, call by call.
Compare with other models
Rates are checkout rates. Full comparison view