Coding
LMArena WebDev Leaderboard
Code Arena (formerly WebDev Arena) is Arena's head-to-head coding leaderboard: two models each build a real web app or UI from the same prompt, and people vote blind on which result is better. The rating shown is a relative Bradley-Terry score, not a percentage — compare it only to other models on this board.
Source: lmarena55 open models ranked+91 proprietaryData through Oct 2026
Open models ranked on LMArena WebDev
# shows rank among open models / rank overall (including proprietary).
| # | Model | Score | VRAM @ Q4 |
|---|---|---|---|
| 1 / 16 | DeepSeek V4.1 Flash 1017 · 763.2B | 1665.0 | ~458 GB |
| 2 / 18 | Kimi K3 · 2779.9B | 1656.4 | ~1674 GB |
| 3 / 22 | MiMo V2.6 Flash RL · 309B | 1637.2 | ~186 GB |
| 4 / 23 | Hy4 Preview · 780.0B | 1631.1 | ~469 GB |
| 5 / 24 | Qwen3.8 Flash Next · 180.0B | 1628.3 | ~108 GB |
| 6 / 25 | MiMo V2.6 Pro RL · 1020B | 1625.4 | ~613 GB |
| 7 / 27 | GLM 5.3 · 753.3B | 1622.8 | ~456 GB |
| 8 / 29 | DeepSeek V4.1 Flash · 763.2B | 1619.2 | ~458 GB |
| 9 / 32 | GLM 5.3 Flash · 321.3B | 1606.6 | ~195 GB |
| 10 / 33 | GLM 5.2 · 753.3B | 1605.6 | ~456 GB |
| 11 / 34 | Qwen3.8 27B · 27.8B | 1589.8 | ~17 GB |
| 12 / 38 | MiniMax M3.1 Flash · 428B | 1582.5 | ~257 GB |
| 13 / 39 | DeepSeek V4 Pro 0813 · 1650.5B | 1581.9 | ~991 GB |
| 14 / 40 | DeepSeek V4 Flash · 290.9B | 1581.8 | ~175 GB |
| 15 / 55 | Muse Spark Open 816B A42B · 822.0B | 1526.2 | ~543 GB |
| 16 / 62 | Kimi K2.6 · 1026.9B | 1509.2 | ~620 GB |
| 17 / 63 | GLM 5.1 · 753.9B | 1507.9 | ~457 GB |
| 18 / 64 | Hy3 · 298.8B | 1505.4 | ~180 GB |
| 19 / 69 | MiniMax M3 · 427.0B | 1483.3 | ~257 GB |
| 20 / 72 | MiMo V2.5 Pro · 1023.2B | 1480.4 | ~614 GB |
| 21 / 73 | Kimi K2.7 Code · 1026.9B | 1473.5 | ~620 GB |
| 22 / 75 | DeepSeek V4 Pro · 1598.8B | 1464.2 | ~960 GB |
| 23 / 83 | MiMo V2.5 · 310.8B | 1438.3 | ~187 GB |
| 24 / 84 | Kimi K2.5 · 1026.9B | 1436.2 | ~620 GB |
| 25 / 85 | Beam 1.0 501B A23B · 501.3B | 1436.0 | ~301 GB |
| 26 / 86 | GLM 5 · 753.9B | 1435.6 | ~457 GB |
| 27 / 87 | GLM 4.7 · 358.3B | 1435.2 | ~216 GB |
| 28 / 92 | Inkling · 952.4B | 1415.2 | ~572 GB |
| 29 / 93 | Inkling Small · 266.0B | 1409.1 | ~160 GB |
| 30 / 97 | Qwen3.5 397B A17B · 403.4B | 1400.2 | ~242 GB |
| 31 / 98 | NVIDIA Nemotron 3.5 Super VL 120B A12B BF16 · 124.5B | 1399.2 | ~82 GB |
| 32 / 99 | MiniMax M2.7 · 228.7B | 1398.2 | ~138 GB |
| 33 / 104 | MiniMax M2.5 · 228.7B | 1386.2 | ~138 GB |
| 34 / 107 | MiniMax M2.1 · 228.7B | 1384.0 | ~138 GB |
| 35 / 110 | Gemma 4 31B IT · 31.3B | 1368.7 | ~20 GB |
| 36 / 111 | DeepSeek V3.2 · 685.4B | 1362.5 | ~415 GB |
| 37 / 112 | Qwen3.5 122B A10B · 125.1B | 1360.4 | ~75 GB |
| 38 / 113 | Gemma 4 26B A4B IT · 25.8B | 1358.8 | ~16 GB |
| 39 / 114 | Qwen3.5 27B · 27.8B | 1358.3 | ~17 GB |
| 40 / 116 | Hy3 Preview · 298.8B | 1356.9 | ~180 GB |
| 41 / 117 | Muse Glimmer 30B · 29.8B | 1355.1 | ~18 GB |
| 42 / 118 | Laguna M.1 · 225.8B | 1349.2 | ~136 GB |
| 43 / 120 | GLM 4.6 · 356.8B | 1341.2 | ~215 GB |
| 44 / 123 | MiMo v2 Flash · 309.8B | 1330.6 | ~186 GB |
| 45 / 125 | Kimi K2 Thinking · 1026.4B | 1321.9 | ~620 GB |
| 46 / 126 | Laguna XS.2 · 33.4B | 1302.4 | ~20 GB |
| 47 / 127 | MiniMax M2 · 228.7B | 1297.1 | ~138 GB |
| 48 / 128 | Qwen3 Coder 480B A35B Instruct · 480.2B | 1274.8 | ~289 GB |
| 49 / 129 | DeepSeek V3.2 Exp · 685.4B | 1271.4 | ~415 GB |
| 50 / 130 | Mistral Medium 3.5 128B · 127.7B | 1264.0 | ~78 GB |
| 51 / 133 | Qwen3.5 35B A3B · 36.0B | 1254.2 | ~22 GB |
| 52 / 137 | Trinity Large Thinking · 398.6B | 1238.0 | ~798 GB |
| 53 / 138 | Mistral Large 3 675B Instruct 2512 · 675B | 1232.6 | ~446 GB |
| 54 / 141 | Devstral 2 123B Instruct 2512 · 125.0B | 1197.1 | ~76 GB |
| 55 / 142 | Granite 4.1 8B · 8.8B | 1192.5 | ~5.9 GB |
Score vs model size
Which models give the most quality for their size — the ones worth running locally.
- Granite 4.1 8B, 9B, score 1192.5 — on the efficiency frontier (best score at its size or smaller).
- Gemma 4 26B A4B IT, 26B, score 1358.8 — on the efficiency frontier (best score at its size or smaller).
- Qwen3.8 27B, 28B, score 1589.8 — on the efficiency frontier (best score at its size or smaller).
- Qwen3.8 Flash Next, 180B, score 1628.3 — on the efficiency frontier (best score at its size or smaller).
- MiMo V2.6 Flash RL, 309B, score 1637.2 — on the efficiency frontier (best score at its size or smaller).
- DeepSeek V4.1 Flash 1017, 763B, score 1665.0 — on the efficiency frontier (best score at its size or smaller).
LMArena WebDev: frequently asked questions
- What is the best open LLM on LMArena WebDev?
- DeepSeek V4.1 Flash 1017 is the top open model on LMArena WebDev, scoring 1665.0. Among all models tested — including proprietary ones — it ranks #16. The top model overall is Claude Opus 5.5 (Max) (Anthropic) at 1809.8.
- What's the best LMArena WebDev model you can run on a 24 GB GPU?
- Qwen3.8 27B is the highest-scoring open model that fits in 24 GB at 4-bit quantization (~17 GB), scoring 1589.8 on LMArena WebDev.
- What's the best LMArena WebDev model you can run on a 12 GB GPU?
- Granite 4.1 8B is the highest-scoring open model that fits in 12 GB at 4-bit quantization (~5.9 GB), scoring 1192.5 on LMArena WebDev.
- Can open models match proprietary models on LMArena WebDev?
- Not quite on LMArena WebDev: the strongest proprietary model (Claude Opus 5.5 (Max)) scores 1809.8, ahead of the best open model (DeepSeek V4.1 Flash 1017) at 1665.0 — but you can run the open one yourself.
Scores aggregated from lmarena. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.