SAGE (Service Agent) - Ecommerce Refund: leaderboard
Metric: Overall Assessment Score (0-100), 0.8 x logical compliance + 0.2 conversational quality, in the Ecommerce Refund scenario (ecommerce refunds (deep decision tree on product status and credit level)), over SAGE customer-service dialogues: standard operating procedures formalized as dialogue graphs, an adversarial simulated user with varied intents and personas, turns 1, 5, 10, 15 and the final turn scored; three unnamed judge LLMs (majority vote and mean) plus a deterministic rule engine produce the reference; closed models through their APIs, open models served with vLLM; 0-100 scale; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 27 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.5 | 72.73 |
| 2 | Qwen 2.5 32B Instruct | 72.07 |
| 3 | DeepSeek V3.2 | 71.51 |
| 4 | Claude Sonnet 4.5 | 71.21 |
| 5 | Gemini 3 Pro (Preview) | 70.99 |
| 6 | GPT-4.1 | 70.32 |
| 7 | Seed 1.8 | 70.25 |
| 8 | DeepSeek R1 | 70.16 |
| 9 | Kimi K2.5 | 69.69 |
| 10 | DeepSeek V3 | 69.62 |
| 11 | Gemini 2.5 Pro | 69.46 |
| 12 | Qwen 2.5 7B Instruct | 69.05 |
| 13 | Qwen 2.5 3B Instruct | 68.92 |
| 14 | Llama 3.3 70B Instruct | 68.6 |
| 15 | MiniMax-M2.1 | 68.33 |
Interactive version: theaggregate.ai/benchmark?slug=sage-service-agent-ecommerce-refund · How It Works · Data refreshed daily, snapshot 2026-10-07.