Design Arena: agents building mobile apps leaderboard
- Claude Opus 5 leads
- 1348
- models measured
- 50
- lead over #2
- +36 Elo
- median
- 1178
Leader on Design Arena: agents building mobile apps · read October 22
On October 22, 2026, 50 of our catalog's models had a Design Arena: agents building mobile apps result, and Anthropic Claude Opus 5 led with 1348, ahead of OpenAI GPT-6 Astra at 1312.
Design Arena shows people two designs made by two models from the same prompt (a mobile app, built by the model as an agent) and asks which is better. The number is the Elo rating built from those votes, as Design Arena publishes it.
Role in the index
Design Arena: agents building mobile apps is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Battles | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5Anthropic | 1348 | Not published | 83.0 | $25.00 | |
| 2 | GPT-6 AstraOpenAI | 1312 | 669 | 85.1 | $50.00 | |
| 3 | Claude Opus 5.5Anthropic | 1308 | 706 | 88.8 | $20.00 | |
| 4 | Kimi K3Moonshot AI | 1269 | 1,164 | 62.3 | $15.00 | |
| 5 | Claude Fable 5.1Anthropic | 1267 | 481 | 84.2 | $50.00 | |
| 6 | Claude Fable 5Anthropic | 1264 | 791 | 79.3 | $50.00 | |
| 7 | Claude Sonnet 5.5Anthropic | 1260 | 299 | 87.3 | $10.00 | |
| 8 | GPT-5.6 SolOpenAI | 1248 | 1,130 | 73.5 | $20.00 | |
| 9 | Gemini 3.8 FlashGoogle | 1236 | 4,252 | 61.6 | $3.75 | |
| 10 | Grok 4.6xAI | 1234 | Not published | 65.7 | $6.00 | |
| 11 | Gemini 3.7 FlashGoogle | 1229 | 683 | 58.0 | $3.75 | |
| 12 | Qwen3.8 Max (0902)Qwen | 1224 | 1,586 | 62.4 | $6.00 | |
| 13 | Claude Opus 4.8Anthropic | 1222 | 1,284 | 65.0 | $25.00 | |
| 14 | Grok 4.5xAI | 1220 | Not published | 58.6 | $6.00 | |
| 15 | Muse Spark 1.3Meta | 1218 | 890 | 57.4 | $4.25 | |
| 16 | GPT-6 SolOpenAI | 1217 | 322 | 76.0 | $10.00 | |
| 17 | Claude Sonnet 5Anthropic | 1215 | 2,010 | 57.7 | $10.00 | |
| 18 | GPT-6.1 SolOpenAI | 1213 | 252 | 82.4 | $10.00 | |
| 19 | Claude Sonnet 4.6Anthropic | 1212 | 1,558 | 57.3 | $15.00 | |
| 20 | Claude Opus 4.6Anthropic | 1207 | 2,190 | 57.2 | $25.00 | |
| 21 | Gemini 3.6 FlashGoogle | 1198 | 1,749 | 62.5 | $3.75 | |
| 22 | GLM 5.3Z.ai | 1194 | 301 | 70.6 | $4.40 | |
| 23 | GPT-5.1-CodexOpenAI | 1191 | Not published | 15.3 | $10.00 | |
| 24 | Claude Opus 4.5Anthropic | 1189 | 1,542 | 44.0 | $25.00 | |
| 25 | Muse Spark 1.2Meta | 1181 | 1,309 | 55.7 | $4.25 | |
| 26 | Muse Spark 1.1Meta | 1175 | Not published | 70.3 | $4.25 | |
| 27 | Gemini 3.5 FlashGoogle | 1174 | 1,373 | 39.8 | $9.00 | |
| 28 | MiniMax M3MiniMax | 1169 | 2,204 | 39.2 | $1.20 | |
| 29 | GLM 5.1Z.ai | 1162 | 3,260 | 49.2 | $4.40 | |
| 30 | GLM 5.2Z.ai | 1159 | 1,879 | 61.1 | $4.40 | |
| 31 | Kimi K2.6Moonshot AI | 1156 | 2,665 | 52.1 | $4.00 | |
| 32 | Qwen3.7 MaxQwen | 1155 | 1,228 | 54.1 | $4.425 | |
| 33 | Kimi K2.5Moonshot AI | 1147 | 1,391 | 38.6 | $2.85 | |
| 34 | Kimi K2.7 CodeMoonshot AI | 1143 | 1,307 | 53.4 | $4.00 | |
| 35 | GPT-5.5OpenAI | 1142 | 843 | 71.3 | $30.00 | |
| 36 | Claude Sonnet 4.5Anthropic | 1137 | 414 | 39.5 | $15.00 | |
| 37 | GLM 5Z.ai | 1136 | 2,854 | 41.7 | $3.20 | |
| 38 | GPT-6 LunaOpenAI | 1134 | 234 | 57.3 | $0.50 | |
| 39 | GLM 5V TurboZ.ai | 1134 | 1,601 | Not benchmarked | $4.00 | |
| 40 | GPT-5.6 TerraOpenAI | 1111 | 1,101 | 63.9 | $12.00 | |
| 41 | GLM 4.7Z.ai | 1105 | 1,550 | 55.1 | $2.20 | |
| 42 | GLM 4.6Z.ai | 1096 | 1,660 | 8.8 | $2.20 | |
| 43 | Gemini 3.1 Pro PreviewGoogle | 1094 | 2,933 | 32.3 | $12.00 | |
| 44 | GPT-5.2OpenAI | 1093 | 1,813 | 55.9 | $14.00 | |
| 45 | Gemini 3 Flash PreviewGoogle | 1090 | 2,015 | 43.7 | $3.00 | |
| 46 | GPT-5.2-CodexOpenAI | 1090 | 1,767 | 45.4 | $14.00 | |
| 47 | Grok 4.20xAI | 1090 | Not published | 33.3 | $2.50 | |
| 48 | GPT-5.4OpenAI | 1087 | 1,929 | 64.4 | $15.00 | |
| 49 | Grok 4.3xAI | 1058 | 1,639 | 37.4 | $2.50 | |
| 50 | GPT-5.3-CodexOpenAI | 1053 | 955 | 60.0 | $14.00 |
Only models in our catalog are listed, and each appears once with its best published run. 145 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out. Every other column is a figure the board publishes for the same run.