Constrained Ad-Copy Variant Generation
Given a product brief, audience, channel limits, and prohibited claims, generate diverse ad variants that satisfy length and compliance rules.
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 | 7% | 100/100 | $31.57 |
| Qwen3 235B A22B | mid | 5% | 100/100 | $74.48 |
| DeepSeek V3 | mid | 19% | 100/100 | $34.25 |
| Mistral Large 2407 | mid | 13% | 100/100 | $83.34 |
| Arcee Trinity Large Thinking | mid | 16% | 100/100 | $81.86 |
| GPT-5.4 | frontier | 83% | 100/100 | $506 |
| Claude Opus 4.7 | frontier | 79% | 100/100 | $1939 |
| Gemini 3.1 Pro Preview | frontier | 40% | 100/100 | $2197 |
| Gemma 4 E4B IT | small | 22% | 100/100 | $4.18 |
| Granite 4.1 8B | small | 5% | 100/100 | $5.16 |
| Ministral 8B Instruct 2410 | small | 5% | 100/100 | $10.70 |
| Qwen3 4B Instruct 2507 | small | 1% | 100/100 | $28.69 |
LLM-judge pass rate on 100 synthetic examples. Generator: gpt-5.2. Judge: gpt-5.2. Evaluated 2026-09-09T03:12:02.527Z.
Directional: measured on a synthetic eval set generated by drydock. Cost/latency are not yet captured for taskrouter-run benchmarks.
drydock