LMArena — AI 模型人类盲测偏好 Elo 排名
LMArena Elo 榜
人类盲测偏好投票：模型有多强（与 OpenRouter/OpenCode 的「用量」互补）
对话之王
claude-fable-5
Elo 1508
编程之王
claude-opus-5-max
Elo 1691
开源 vs 闭源
20 : 80
对话+编程双榜模型数
总投票数
227.0万
覆盖 100 个模型
| # | 模型 | 协议 | Elo | 95% CI | 投票 |
|---|---|---|---|---|---|
| 1 | claude-fable-5 Anthropic | proprietary | 1508 | ±5 | 2.4万 |
| 2 | claude-opus-4-6-high Anthropic | proprietary | 1504 | ±4 | 7.2万 |
| 3 | claude-opus-4-7-high Anthropic | proprietary | 1502 | ±4 | 6.0万 |
| 4 | muse-spark-1.2 (xHigh) Meta | proprietary | 1498 | ±10 | 3.3k |
| 5 | claude-opus-4-6 Anthropic | proprietary | 1497 | ±3 | 7.6万 |
| 6 | claude-opus-4-7 Anthropic | proprietary | 1494 | ±4 | 6.1万 |
| 7 | claude-opus-5-high Anthropic | proprietary | 1493 | ±5 | 2.8万 |
| 8 | muse-spark-1.1 Meta | proprietary | 1491 | ±5 | 2.0万 |
| 9 | gemini-3.7-flash-high Google | proprietary | 1490 | ±8 | 5.7k |
| 10 | kimi-k3-max Moonshot | — | 1489 | ±6 | 1.5万 |
| 11 | muse-spark Meta | proprietary | 1488 | ±6 | 1.4万 |
| 12 | claude-opus-5-max Anthropic | proprietary | 1487 | ±6 | 1.3万 |
| 13 | glm-5.3-max Z.ai | open | 1487 | ±10 | 3.8k |
| 14 | gemini-3.1-pro-preview Google | proprietary | 1486 | ±3 | 9.9万 |
| 15 | gemini-3-pro Google | proprietary | 1485 | ±4 | 4.1万 |
| 16 | gpt-5.6-sol-xhigh OpenAI | proprietary | 1482 | ±5 | 2.0万 |
| 17 | gpt-5.5-high OpenAI | proprietary | 1482 | ±4 | 6.0万 |
| 18 | claude-opus-4-8-high Anthropic | proprietary | 1482 | ±4 | 4.5万 |
| 19 | qwen3.8-max Alibaba | proprietary | 1481 | ±7 | 10.0k |
| 20 | gemini-3.6-flash-high Google | proprietary | 1481 | ±5 | 1.8万 |
| 21 | gemini-3.5-flash-high Google | proprietary | 1478 | ±5 | 3.0万 |
| 22 | gpt-5.4-high OpenAI | proprietary | 1476 | ±4 | 6.1万 |
| 23 | gpt-5.5 OpenAI | proprietary | 1476 | ±4 | 6.1万 |
| 24 | gpt-5.2-chat-latest-20260210 OpenAI | proprietary | 1476 | ±4 | 3.4万 |
| 25 | grok-4.20-beta1 SpaceXAI | proprietary | 1475 | ±5 | 2.7万 |
| 26 | gemini-3.5-flash-medium Google | proprietary | 1474 | ±5 | 2.8万 |
| 27 | qwen3.7-max-preview Alibaba | proprietary | 1474 | ±10 | 3.7k |
| 28 | gpt-5.5-instant OpenAI | proprietary | 1474 | ±5 | 2.6万 |
| 29 | claude-opus-4-8 Anthropic | proprietary | 1473 | ±4 | 4.5万 |
| 30 | gemini-3-flash Google | proprietary | 1473 | ±4 | 3.1万 |
| 31 | claude-opus-4-5-20251101-high-32k Anthropic | proprietary | 1473 | ±4 | 3.7万 |
| 32 | claude-sonnet-4-6 Anthropic | proprietary | 1472 | ±4 | 6.7万 |
| 33 | grok-4.20-beta-0309-reasoning SpaceXAI | proprietary | 1472 | ±4 | 6.2万 |
| 34 | glm-5.2-max Z.ai | open | 1470 | ±5 | 3.1万 |
| 35 | grok-4.5 SpaceXAI | proprietary | 1470 | ±5 | 2.2万 |
| 36 | grok-4.20-multi-agent-beta-0309 SpaceXAI | proprietary | 1470 | ±4 | 6.1万 |
| 37 | claude-opus-4-5-20251101 Anthropic | proprietary | 1469 | ±3 | 7.1万 |
| 38 | ernie-5.1 Baidu | proprietary | 1468 | ±5 | 3.7万 |
| 39 | glm-5.1 Z.ai | open | 1468 | ±4 | 4.2万 |
| 40 | mimo-v2.5-pro Xiaomi | open | 1468 | ±4 | 5.4万 |
| 41 | gpt-5.4 OpenAI | proprietary | 1466 | ±4 | 6.4万 |
| 42 | grok-4.1-thinking SpaceXAI | proprietary | 1466 | ±3 | 6.6万 |
| 43 | qwen3.5-max-preview Alibaba | proprietary | 1465 | ±5 | 2.1万 |
| 44 | gpt-5.6-terra-xhigh OpenAI | proprietary | 1465 | ±5 | 2.0万 |
| 45 | claude-sonnet-5-high Anthropic | proprietary | 1461 | ±5 | 2.7万 |
| 46 | grok-4.6-high SpaceXAI | proprietary | 1461 | ±10 | 3.5k |
| 47 | kimi-k2.6 Moonshot | open | 1461 | ±5 | 3.8万 |
| 48 | qwen3.6-max-preview Alibaba | proprietary | 1460 | ±8 | 5.2k |
| 49 | grok-4.1 SpaceXAI | proprietary | 1459 | ±3 | 6.8万 |
| 50 | deepseek-v4-pro-high-20260813 DeepSeek | open | 1459 | ±10 | 3.6k |