DesignArena: Python-PPTX Slides

Two agents each write python-pptx code that builds a PowerPoint deck from the same real user prompt. A person compares the two decks on narrative structure, visual design, and presentation quality and votes for one, and a Bradley–Terry fit over every vote rates each model on an Elo scale that starts at 1200. Cost is each model’s list price per million tokens, not spend per deck.

Last updated · Source leaderboard checked daily

Main Insights

Results

DesignArena: Python-PPTX Slides results, ranked by score (elo), highest first. Last updated .
RankModelProviderScore (Elo)Cost (USD / task)
1Claude Opus 5.5 (high)Anthropic1390.0%$8.00
2GPT-6 Astra (max)OpenAI1368.0%$20.00
3Claude Fable 5.1Anthropic1316.0%$20.00
4GPT-6.1 Sol (max)OpenAI1304.0%$4.00
5Kimi K3.1 ProMoonshot AI1278.0%$4.00
6Claude Fable 5Anthropic1266.0%$20.00
7Qwen 3.8 MaxAlibaba1251.0%$3.00
8Inkling 2 SmallThinking Machines1244.0%$0.47
9Claude Opus 5Anthropic1225.0%$10.00
10Kimi K3Moonshot AI1223.0%$6.00
11GLM 5.3Z.ai1217.0%$2.15
12Grok 4.6xAI1213.0%$3.00
13Muse Spark 1.3 (xhigh)Meta1212.0%$2.00
14GPT-5.6 Sol (medium)OpenAI1205.0%$8.00
15Grok 4.5xAI1198.0%$3.00