Multiple-Choice Distractor Generation
Given a question, correct answer, and misconception taxonomy, generate plausible distinct distractors tied to specified misconceptions.
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 | 39% | 100/100 | $35.79 |
| Qwen3 235B A22B | mid | 52% | 100/100 | $90.82 |
| DeepSeek V3 | mid | 60% | 100/100 | $20.36 |
| Mistral Large 2407 | mid | 51% | 100/100 | $53.65 |
| Arcee Trinity Large Thinking | mid | 15% | 100/100 | $153 |
| GPT-5.4 | frontier | 71% | 100/100 | $246 |
| Claude Opus 4.7 | frontier | 69% | 100/100 | $1353 |
| Gemini 3.1 Pro Preview | frontier | 60% | 100/100 | $1518 |
| Gemma 4 E4B IT | small | 36% | 100/100 | $2.53 |
| Granite 4.1 8B | small | 31% | 100/100 | $6.08 |
| Ministral 8B Instruct 2410 | small | 23% | 100/100 | $7.37 |
| Qwen3 4B Instruct 2507 | small | 18% | 100/100 | $46.72 |
LLM-judge pass rate on 100 synthetic examples. Generator: gpt-5.2. Judge: gpt-5.2. Evaluated 2026-09-09T09:06:02.690Z.
Directional: measured on a synthetic eval set generated by drydock. Cost/latency are not yet captured for taskrouter-run benchmarks.
drydock