Design Arena: agents building web apps leaderboard
- GPT-6 Astra leads
- 1373
- models measured
- 50
- lead over #2
- +55 Elo
- median
- 1204
Leader on Design Arena: agents building web apps · read October 22
On October 22, 2026, 50 of our catalog's models had a Design Arena: agents building web apps result, and OpenAI GPT-6 Astra led with 1373, ahead of Anthropic Claude Fable 5.1 at 1318.
Design Arena shows people two designs made by two models from the same prompt (a web 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 web 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 | GPT-6 AstraOpenAI | 1373 | 720 | 85.1 | $50.00 | |
| 2 | Claude Fable 5.1Anthropic | 1318 | 745 | 84.2 | $50.00 | |
| 3 | Claude Opus 5.5Anthropic | 1316 | 630 | 88.8 | $20.00 | |
| 4 | GPT-6.1 SolOpenAI | 1306 | 501 | 82.4 | $10.00 | |
| 5 | Claude Sonnet 5.5Anthropic | 1305 | 517 | 87.3 | $10.00 | |
| 6 | Qwen3.8 Max (0902)Qwen | 1300 | 1,784 | 62.4 | $6.00 | |
| 7 | Muse Spark 1.3Meta | 1297 | 5,141 | 57.4 | $4.25 | |
| 8 | Kimi K3Moonshot AI | 1295 | 1,971 | 62.3 | $15.00 | |
| 9 | MiMo-V2.6-ProXiaomi | 1290 | Not published | 65.4 | $0.87 | |
| 10 | Kimi K2.6Moonshot AI | 1268 | Not published | 52.1 | $4.00 | |
| 11 | Claude Opus 5Anthropic | 1266 | 1,916 | 83.0 | $25.00 | |
| 12 | Claude Fable 5Anthropic | 1261 | 2,911 | 79.3 | $50.00 | |
| 13 | Grok 4.6xAI | 1248 | Not published | 65.7 | $6.00 | |
| 14 | Claude Sonnet 5Anthropic | 1245 | 4,684 | 57.7 | $10.00 | |
| 15 | GPT-5.6 SolOpenAI | 1245 | 2,597 | 73.5 | $20.00 | |
| 16 | Gemini 3.8 FlashGoogle | 1238 | 1,500 | 61.6 | $3.75 | |
| 17 | Claude Opus 4.8Anthropic | 1233 | 3,711 | 65.0 | $25.00 | |
| 18 | GPT-6 SolOpenAI | 1231 | 696 | 76.0 | $10.00 | |
| 19 | Gemini 3.7 FlashGoogle | 1223 | 1,309 | 58.0 | $3.75 | |
| 20 | GLM 5.2Z.ai | 1222 | 6,333 | 61.1 | $4.40 | |
| 21 | Muse Spark 1.2Meta | 1221 | 2,092 | 55.7 | $4.25 | |
| 22 | Grok 4.5xAI | 1218 | Not published | 58.6 | $6.00 | |
| 23 | Qwen3.7 MaxQwen | 1208 | 2,981 | 54.1 | $4.425 | |
| 24 | Muse Spark 1.1Meta | 1207 | Not published | 70.3 | $4.25 | |
| 25 | Claude Opus 4.6Anthropic | 1206 | 4,192 | 57.2 | $25.00 | |
| 26 | Gemini 3.5 FlashGoogle | 1202 | 5,745 | 39.8 | $9.00 | |
| 27 | MiniMax M3MiniMax | 1200 | 3,647 | 39.2 | $1.20 | |
| 28 | Gemini 3.6 FlashGoogle | 1197 | 2,871 | 62.5 | $3.75 | |
| 29 | T InklingThinking Machines | 1196 | 1,207 | 31.5 | $4.05 | |
| 30 | GPT-6 LunaOpenAI | 1191 | 647 | 57.3 | $0.50 | |
| 31 | Claude Sonnet 4.6Anthropic | 1190 | 2,825 | 57.3 | $15.00 | |
| 32 | Kimi K2.7 CodeMoonshot AI | 1176 | 4,373 | 53.4 | $4.00 | |
| 33 | GLM 5.1Z.ai | 1175 | 7,186 | 49.2 | $4.40 | |
| 34 | Claude Opus 4.5Anthropic | 1159 | 2,697 | 44.0 | $25.00 | |
| 35 | Grok 4.20xAI | 1144 | Not published | 33.3 | $2.50 | |
| 36 | Kimi K2.5Moonshot AI | 1136 | 3,435 | 38.6 | $2.85 | |
| 37 | Grok 4.3xAI | 1126 | 6,464 | 37.4 | $2.50 | |
| 38 | Gemini 3.1 Pro PreviewGoogle | 1125 | 6,649 | 32.3 | $12.00 | |
| 39 | GLM 5V TurboZ.ai | 1124 | 5,385 | Not benchmarked | $4.00 | |
| 40 | GPT-5.6 TerraOpenAI | 1120 | 3,548 | 63.9 | $12.00 | |
| 41 | GPT-5.5OpenAI | 1117 | 5,342 | 71.3 | $30.00 | |
| 42 | Gemini 3 Flash PreviewGoogle | 1111 | 3,979 | 43.7 | $3.00 | |
| 43 | Solar Pro 4Upstage | 1105 | 1,315 | Not benchmarked | $0.60 | |
| 44 | GPT-5.2OpenAI | 1086 | 3,570 | 55.9 | $14.00 | |
| 45 | Claude Sonnet 4.5Anthropic | 1060 | 1,503 | 39.5 | $15.00 | |
| 46 | GPT-5.4OpenAI | 1059 | 3,320 | 64.4 | $15.00 | |
| 47 | GPT-5.2-CodexOpenAI | 1051 | 3,426 | 45.4 | $14.00 | |
| 48 | GPT-5.3-CodexOpenAI | 1032 | 2,245 | 60.0 | $14.00 | |
| 49 | GPT-5.1-CodexOpenAI | 1018 | 525 | 15.3 | $10.00 | |
| 50 | DeepSeek V4 Pro 0423DeepSeek | 1000 | Not published | 38.7 | $2.976 |
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.