SAGE (Service Agent): leaderboard
Metric: Overall Assessment Score (0-100), 0.8 x logical compliance + 0.2 conversational quality, averaged over the six scenarios, 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 August 2027. 27 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.5 | 72.62 |
| 2 | Claude Sonnet 4.5 | 71.56 |
| 3 | Gemini 3 Pro (Preview) | 71.4 |
| 4 | DeepSeek V3.2 | 71.29 |
| 5 | Kimi K2.5 | 68.71 |
| 6 | DeepSeek R1 | 68.39 |
| 7 | MiniMax-M2.1 | 68.33 |
| 8 | Gemini 3 Flash (Preview) | 68.28 |
| 9 | GLM-4.7 | 67.25 |
| 10 | Gemini 2.5 Pro | 66.9 |
| 11 | GPT-4.1 | 66.79 |
| 12 | DeepSeek V3 | 66.54 |
| 13 | Seed 1.8 | 66.14 |
| 14 | Qwen 2.5 32B Instruct | 65.34 |
| 15 | Qwen 2.5 72B Instruct | 65.27 |
Interactive version: theaggregate.ai/benchmark?slug=sage-service-agent · How It Works · Data refreshed daily, snapshot 2026-10-07.