DeepSeekDeepSeek V4 Pro
DeepSeek's V4-generation flagship, succeeding V3.2 — which held the highest checkout rate in our prior benchmark (68%) with an autonomous, push-through-errors style. Newly added to the lineup; playground benchmarking is in progress, so no V4 completion metrics yet.
Avg Tokens113,049Avg Duration65.7sTurns to Checkout15.8
Shopping Score19/100
Weak
Scores computed from real agent sessions against live UCP-enabled stores. Not estimated — every data point is from an actual tool call, checkout attempt, and store response.
Shopping Score Breakdown
Checkout Rate
Cart Rate
Search Rate
Turn Efficiency
Token Efficiency
22.4% of sessions failed with errors.
Top Stores
| Store | Checkout % | Cart % |
|---|---|---|
| •••••••••••••••••••••••••••••••• | 50% | 50% |
| ••••••••••••••••••••••••••••• | 9.5% | 57.1% |
| ••••••••••••••••••••••••••••• | 0% | 50% |
| everlane.com | 0% | 16.7% |
| singon.com.np | 0% | 0% |
| allbirds.com | 0% | 0% |
| •••••••••••••••••••••••••••••••• | 0% | 66.7% |
| ••••••••••••••••• | 0% | 50% |
Known Issues
No known issues documented for this model yet.
Token Usage
Avg per Session113,049
Fleet Average66,097
vs Fleet+71%
Median (p50)49,426
p90273,528
Prompt / Completion split
Prompt 111,048 (98%) Completion 2,001 (2%)
Daily Average Tokens (last 30 days)
Distribution range
Min2,711p2521,203p5049,426p75146,575p90273,528Max666,190
Compare With Other Models
GoogleGemini 3.5 Flash39.6%
MetaLlama 3.3 70B39.3%
AnthropicClaude Sonnet 4.636.9%GoogleGemini 2.5 Flash26.5%
xAIGrok 4.326.3%AnthropicClaude Opus 4.823.3%DeepSeekDeepSeek V3.222.1%GoogleGemini 3.1 Pro21.4%GoogleGemini 2.5 Pro20.1%
OpenAIGPT-4o15.8%AnthropicClaude Sonnet 511.4%OpenAIGPT-5.511.3%xAIGrok 3 Mini (Reasoning)11.1%OpenAIo4-mini (Reasoning)9.7%DeepSeekDeepSeek V4 Pro9%
Moonshot AIKimi K35%xAIGrok 4.54.5%DeepSeekDeepSeek R1 (Reasoning)4.1%DeepSeekDeepSeek V4 Flash3.4%MetaLlama 4 Maverick2.8%AnthropicClaude Opus 52%GoogleGemini 3.6 Flash2%
AlibabaQwen 3.7 Max0%AlibabaQwQ 32B (Reasoning)0%