SAGE (Service Agent) - Online Education: leaderboard
Metric: Overall Assessment Score (0-100), 0.8 x logical compliance + 0.2 conversational quality, in the Online Education scenario (online education contract disputes and refund-risk control), 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 January 2027. 27 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 82.15 |
| 2 | DeepSeek V3.2 | 78.17 |
| 3 | Gemini 3 Flash (Preview) | 75.62 |
| 4 | Claude Opus 4.5 | 75.22 |
| 5 | Claude Sonnet 4.5 | 74.24 |
| 6 | MiniMax-M2.1 | 72.55 |
| 7 | GLM-4.7 | 72.36 |
| 8 | GPT-4.1 | 71.69 |
| 9 | Kimi K2.5 | 71.15 |
| 10 | Gemini 2.5 Pro | 70.28 |
| 11 | DeepSeek R1 | 69.4 |
| 12 | DeepSeek V3 | 68.77 |
| 13 | Qwen 2.5 72B Instruct | 68.76 |
| 14 | Qwen 3 235B A22B | 68.02 |
| 15 | Qwen 3 32B | 67.56 |
Interactive version: theaggregate.ai/benchmark?slug=sage-service-agent-online-education · How It Works · Data refreshed daily, snapshot 2026-10-07.