Warehouse Near-Miss Theme Summary
Given safety logs, classify incident type and contributing factors, merge duplicates, and summarize recurring prevention opportunities.
View token rates · observed 9/11/2026
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 | 3% | 100/100 | $32.55 |
| Qwen3 235B A22B | mid | 0% | 100/100 | $198 |
| DeepSeek V3 | mid | 34% | 100/100 | $80.37 |
| Mistral Large 2407 | mid | 24% | 100/100 | $175 |
| Arcee Trinity Large Thinking | mid | 14% | 100/100 | $110 |
| GPT-5.4 | frontier | 80% | 100/100 | $1199 |
| Claude Opus 4.7 | frontier | 31% | 100/100 | $3880 |
| Gemini 3.1 Pro Preview | frontier | 24% | 100/100 | $2518 |
| Gemma 4 E4B IT | small | 51% | 100/100 | $6.28 |
| Granite 4.1 8B | small | 4% | 100/100 | $10.60 |
| Ministral 8B Instruct 2410 | small | 2% | 100/100 | $24.51 |
| Qwen3 4B Instruct 2507 | small | 1% | 100/100 | $68.23 |
LLM-judge pass rate on 100 synthetic examples. Generator: gpt-5.2. Judge: gpt-5.2. Evaluated 2026-09-09T10:53:02.469Z.
Directional: measured on a synthetic eval set generated by drydock. Cost/latency are not yet captured for taskrouter-run benchmarks.
drydock