SAGE (Service Agent) - Telecom Package: leaderboard
Metric: Overall Assessment Score (0-100), 0.8 x logical compliance + 0.2 conversational quality, in the Telecom Package scenario (telecom package billing and plan upgrades (linear SOP)), 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 Sonnet 4.5 | 75.03 |
| 2 | Claude Opus 4.5 | 73.12 |
| 3 | DeepSeek V3 | 71.6 |
| 4 | DeepSeek V3.2 | 70.61 |
| 5 | DeepSeek R1 | 70.39 |
| 6 | MiniMax-M2.1 | 70.12 |
| 7 | Kimi K2.5 | 69.41 |
| 8 | Qwen 2.5 72B Instruct | 69.16 |
| 9 | Gemini 3 Pro (Preview) | 69.09 |
| 10 | Qwen 2.5 32B Instruct | 68.27 |
| 11 | Qwen 3 14B | 68.01 |
| 12 | GPT-4.1 | 66.9 |
| 13 | Gemini 2.5 Pro | 66.42 |
| 14 | GLM-4.7 | 66.02 |
| 15 | Seed 1.8 | 65.89 |
Interactive version: theaggregate.ai/benchmark?slug=sage-service-agent-telecom-package · How It Works · Data refreshed daily, snapshot 2026-10-07.