SAGE (Service Agent) - Logical Compliance: leaderboard
SAGE logical compliance: whether a customer-service agent classifies the user state, plans the procedure path and takes the action that the dialogue graph and a rule engine prescribe. 0-100 scale, higher is better. It is unrelated to the SAGE robustness benchmark.
Metric: Logical Compliance Score (0-100): 0.4 classification accuracy + 0.4 path correctness + 0.2 action correctness against the graph, 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 March 2027. 27 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 | 72.1 | #165 |
| 2 | DeepSeek V3.2 | 72.08 | #425 |
| 3 | Gemini 3 Pro (Preview) | 71.27 | #123 |
| 4 | Claude Sonnet 4.5 | 70.1 | #267 |
| 5 | DeepSeek R1 | 70 | #484 |
| 6 | Kimi K2.5 | 69.16 | #272 |
| 7 | GLM-4.7 | 68.68 | #383 |
| 8 | GPT-4.1 | 68.31 | #506 |
| 9 | Gemini 3 Flash (Preview) | 68.16 | #156 |
| 10 | DeepSeek V3 | 67.9 | #736 |
| 11 | Qwen Max (SAGE checkpoint unspecified) | 67.57 | |
| 12 | MiniMax-M2.1 | 66.59 | #582 |
| 13 | Qwen 2.5 32B Instruct | 66.49 | #1253 |
| 14 | Llama 3.3 70B Instruct | 66.06 | #1394 |
| 15 | Qwen 3 32B | 65.97 | #983 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=sage-service-agent-logical-compliance · How It Works · Data refreshed daily, snapshot 2026-11-22.