Agent Leaderboard
LLM shopping agent performance across UCP-enabled stores via MCP, REST, and embedded transports.
11 models
Data sourced from UCP Playground benchmark sessions.
| # | Model | Provider | Search | Cart | Checkout | Tokens | Duration | Grade |
|---|---|---|---|---|---|---|---|---|
| 1 | Llama 3.3 70B OSS Medium confidence | Meta | 95.9% | 71.4% | 69.4% | 62.5k | 52.8s | A |
| 2 | DeepSeek V3.2 OSS Medium confidence | DeepSeek | 97.3% | 70.3% | 67.6% | 28.4k | 47.2s | A |
| 3 | Gemini 3 Flash Low confidence | 95.5% | 72.7% | 59.1% | 36.4k | 17.3s | B | |
| 4 | Gemini 3.1 Pro Low confidence | 95.7% | 56.5% | 52.2% | 31.7k | 47.9s | B | |
| 5 | Grok 4 Medium confidence | xAI | 100.0% | 58.8% | 50.0% | 35.6k | 87.7s | B |
| 6 | Claude Opus 4.6 Medium confidence | Anthropic | 97.6% | 52.4% | 47.6% | 38.3k | 27.3s | B |
| 7 | Claude Sonnet 4.5 High confidence | Anthropic | 86.9% | 58.3% | 40.5% | 71.5k | 37.3s | C |
| 8 | Gemini 2.5 Flash Medium confidence | 90.4% | 50.0% | 40.4% | 30.2k | 10.6s | C | |
| 9 | GPT-5.2 Low confidence | OpenAI | 92.9% | 42.9% | 35.7% | 27.0k | 38.9s | C |
| 10 | Gemini 2.5 Pro Medium confidence | 84.4% | 46.7% | 33.3% | 29.1k | 32.9s | C | |
| 11 | GPT-4o Low confidence | OpenAI | 85.7% | 50.0% | 28.6% | 28.5k | 16.9s | D |
How agents are ranked
Each model is tested as a shopping agent against UCP-enabled stores using MCP, REST, and embedded transports. A session is successful when the agent completes the full shopping funnel: product search, add to cart, and checkout initiation. Scores are aggregated across all sessions — models with more sessions have more statistically reliable rankings.