Customer Health Signal Classification
Given synthetic usage, support, renewal, and satisfaction signals, assign a health tier and return the observed factors that drove the classification.
View token rates
Compare 2 router rates · observed 9/11/2026
Compare 2 router rates · observed 9/11/2026
All 12 model results and methodology
Generated 100/100 examples
| Model | Tier | Quality | Judged | Scenario cost |
|---|---|---|---|---|
| Nemotron Nano 9B v2 | small | 48% | 100/100 | $15.48 |
| Qwen3 235B A22B | mid | 69% | 100/100 | $24.84 |
| DeepSeek V3 | mid | 64% | 100/100 | $13.65 |
| Mistral Large 2407 | mid | 51% | 100/100 | $30.23 |
| Arcee Trinity Large Thinking | mid | 52% | 100/100 | $84.46 |
| GPT-5.4 | frontier | 70% | 100/100 | $234 |
| Claude Opus 4.7 | frontier | 68% | 100/100 | $620 |
| Gemini 3.1 Pro Preview | frontier | 64% | 100/100 | $865 |
| Gemma 4 E4B IT | small | 62% | 100/100 | $1.15 |
| Granite 4.1 8B | small | 68% | 100/100 | $2.25 |
| Ministral 8B Instruct 2410 | small | 48% | 100/100 | $7.11 |
| Qwen3 4B Instruct 2507 | small | 69% | 100/100 | $12.33 |
LLM-judge pass rate on 100 synthetic examples. Generator: gpt-5.2. Judge: gpt-5.2. Evaluated 2026-09-09T03:33:02.632Z.
Directional: measured on a synthetic eval set generated by drydock. Cost/latency are not yet captured for taskrouter-run benchmarks.
drydock