Customer Support BPM on Customer Support BPM Dataset (test)
79.15Tool Accuracypolicy–critic pipeline
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| policy–critic pipelineEvaluation Protocol=domain-specific SFT2026.04 | 79.15 | 65.48 | |
| Gemini 3 ProEvaluation Protocol=zero-shot prompting2026.04 | 47.78 | 32.93 | |
| Qwen3-235B-A22BEvaluation Protocol=zero-shot prompting2026.04 | 44.33 | 26.18 | |
| GPT-5.2Evaluation Protocol=zero-shot prompting2026.04 | 43.01 | 27.02 | |
| DeepSeek V3.1Evaluation Protocol=zero-shot prompting2026.04 | 42.4 | 27.83 | |
| Qwen2.5-Instruct-7B (baseline)Evaluation Protocol=zero-shot prompting2026.04 | 34.95 | 10.27 |