Ad Writing Arena
Frontier LLMs write a short video ad script from the same brief. A fixed panel of LLM judges compares the scripts in blind pairs. A Bradley–Terry model turns the verdicts into ratings. The product is a real car perfume; its brand name and trademarks are changed on this page.
Leaderboard
| # | Model | Lab | Setting | Rating | 95% interval | Top-5 chance | Win rate | Constraints met | Exact length | Cost / script | Time / script | Judgments |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | Anthropic | low reasoning | 1365 | 1316–1455 | 100% | 79% | 82% | 3/3 | $0.0486 | 16 s | 234 |
| 2 | Muse Wonder | Meta | low reasoning | 1224 | 1136–1316 | 96% | 67% | 83% | 3/3 | $0.0380 | 10 s | 232 |
| 3 | Kimi K3.1 Pro | Moonshot AI | low reasoning | 1210 | 1140–1288 | 96% | 63% | 80% | 3/3 | $0.0328 | 71 s | 234 |
| 4 | Grok 4.7 | xAI | low reasoning | 1191 | 1122–1294 | 86% | 68% | 95% | 2/3 | $0.0190 | 22 s | 232 |
| 5 | MiMo V2.6 Pro | Xiaomi | low reasoning | 1142 | 1064–1226 | 54% | 48% | 54% | 1/3 | $0.0030 | 41 s | 231 |
| 6 | DeepSeek V4 Pro | DeepSeek | low reasoning | 1139 | 1049–1250 | 50% | 59% | 79% | 3/3 | $0.0063 | 43 s | 233 |
| 7 | GPT-6 Astra | OpenAI | low reasoning | 1099 | 1039–1177 | 13% | 60% | 100% | 3/3 | n/a | 31 s | 232 |
| 8 | GPT-6.1 Sol | OpenAI | low reasoning | 1072 | 1005–1146 | 4% | 56% | 95% | 3/3 | n/a | 22 s | 233 |
| 9 | Gemini 4 Argon | low reasoning | 995 | 882–1099 | 0% | 27% | 33% | 0/3 | $0.0123 | 11 s | 234 | |
| 10 | GLM-5.3 Prime | Z.ai | low reasoning | 917 | 805–1020 | 0% | 19% | 24% | 1/3 | $0.0124 | 7 s | 233 |
| 11 | Qwen 3.8 Max | Alibaba | low reasoning | 728 | 640–790 | 0% | 9% | 13% | 1/3 | $0.0472 | 31 s | 233 |
| 12 | Claude Fable 5.1 | Anthropic | low reasoning | 725 | 597–824 | 0% | 6% | 6% | 0/3 | $0.0925 | 14 s | 234 |
| 13 | Llama 3.1 70B anchor | Meta, 2024 | no reasoning | 697 | 584–778 | 0% | 8% | 24% | 1/3 | $0.0019 | 9 s | 234 |
| 14 | Mistral Small 3 anchor | Mistral, early 2025 | no reasoning | 497 | 375–571 | 0% | 0% | 4% | 3/3 | $0.0002 | 13 s | 233 |
Replay
Choose a brief and play the round. Each writer returns one script. Judges compare the scripts in pairs. A list ranks this brief by win rate after each judgment. Pause, drag the timeline, or step one judgment at a time. When the round is paused or finished, select a model to read its script.
Winning scripts
One script per brief: the highest win rate on that brief.
Losing scripts
One script per brief: the lowest win rate on that brief.
Data
Every script and every verdict is in this section. One JSON file holds every brief, prompt, script and judgment, with the brand name changed. The methodology has the rubric and the judge prompt.
All scripts and verdicts
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 |
| round1.json | Judgments of the first judge panel, replaced on 30 Oct 2026 |
| 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 |