Benchmark · Advertising copy
Copywriting Arena
AI models write a short video ad script from the same brief. A fixed panel of AI judges compares the scripts in blind pairs. A Bradley–Terry model turns the verdicts into ratings. The product is a real Indian car perfume; its brand name and trademarks are changed on this page.
Leaderboard
| # | Model | Lab | Setting | Rating | 95% interval | Top-5 chance | Win rate | Gates passed | Exact length | Cost / script | Time / script | Judgments |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | Anthropic | low reasoning | 1389 | 1327–1482 | 100% | 89% | 91% | 3/3 | $0.0486 | 16 s | 234 |
| 2 | Muse Wonder | Meta | low reasoning | 1177 | 1105–1260 | 93% | 68% | 93% | 3/3 | $0.0380 | 10 s | 233 |
| 3 | Kimi K3.1 Pro | Moonshot AI | low reasoning | 1176 | 1121–1264 | 91% | 68% | 89% | 3/3 | $0.0328 | 71 s | 233 |
| 4 | MiMo V2.6 Pro | Xiaomi | low reasoning | 1157 | 1075–1235 | 79% | 57% | 68% | 1/3 | $0.0030 | 41 s | 231 |
| 5 | Grok 4.7 | xAI | low reasoning | 1153 | 1085–1227 | 80% | 66% | 96% | 2/3 | $0.0190 | 22 s | 234 |
| 6 | GPT-6 Astra | OpenAI | low reasoning | 1125 | 1062–1202 | 48% | 64% | 100% | 3/3 | n/a | 31 s | 234 |
| 7 | DeepSeek V4 Pro | DeepSeek | low reasoning | 1067 | 999–1148 | 7% | 50% | 80% | 3/3 | $0.0063 | 43 s | 233 |
| 8 | GPT-6.1 Sol | OpenAI | low reasoning | 1039 | 963–1114 | 2% | 54% | 98% | 3/3 | n/a | 22 s | 233 |
| 9 | Gemini 4 Argon | low reasoning | 975 | 856–1082 | 1% | 28% | 34% | 0/3 | $0.0123 | 11 s | 232 | |
| 10 | GLM-5.3 Prime | Z.ai | low reasoning | 899 | 790–993 | 0% | 22% | 26% | 1/3 | $0.0124 | 7 s | 231 |
| 11 | Claude Fable 5.1 | Anthropic | low reasoning | 799 | 706–884 | 0% | 12% | 15% | 0/3 | $0.0925 | 14 s | 232 |
| 12 | Qwen 3.8 Max | Alibaba | low reasoning | 741 | 646–815 | 0% | 8% | 12% | 1/3 | $0.0472 | 31 s | 234 |
| 13 | Llama 3.1 70B anchor | Meta, 2024 | no reasoning | 675 | 560–762 | 0% | 7% | 25% | 1/3 | $0.0019 | 9 s | 233 |
| 14 | Mistral Small 3 anchor | Mistral, early 2025 | no reasoning | 628 | 494–722 | 0% | 5% | 10% | 3/3 | $0.0002 | 13 s | 231 |
Replay
Select a brief and play the round. The writers receive the brief, each writer returns a script, and then the judges compare the scripts in pairs. You can pause, drag the timeline, or step through the judgments one at a time.
Scripts
Read each script in full, or compare two scripts with every judge's verdict on that pair. To read without knowing the author, hide the model names.
Data
One JSON file holds every brief, prompt, script and judgment shown on this page, with the brand name changed. The methodology has the rubric and the judge prompt.
Product facts (the public part of the brand dossier)
Writer system prompt
Writer prompt for the selected brief
For agents
Every page has a plain-text version. Start with llms.txt.
| llms.txt | Index of the machine-readable files |
| index.md | This page as Markdown: leaderboard, briefs, definitions |
| methodology.md | The full methodology and rubric |
| briefs/B1.md, briefs/B2.md, briefs/B3.md | One Markdown file per brief, with every script in full |
| leaderboard.json | Ratings, intervals and metrics per model |
| data.json | Everything: briefs, prompts, scripts, judgments |