---
title: "指令遵循排行榜"
description: "按格式、长度、步骤等约束作答的排名。"
canonical: https://pirouter.ai/zh/rankings/instruct
last-updated: 2026-10-15
---

# 指令遵循排行榜

> 按格式、长度、步骤等约束作答的排名。

- 更新于: 2026-09-30
- 3,713,332 票 · 385 个模型
- 分数 · 越高越好

## 完整排名

| Rank | Rank spread | Model | Vendor | Score | Votes |
|---|---|---|---|---|---|
| 1 | 1–4 | Gemini 4 Argon (high) | Google | 1529 ±14 | 1808 |
| 2 | 1–7 | [Claude Opus 5.5 \(high\)](https://pirouter.ai/zh/models/anthropic/claude-opus-5.5) | Anthropic | 1514 ±16 | 1337 |
| 3 | 1–4 | Claude Opus 4.6 (high) | Anthropic | 1514 ±5 | 24809 |
| 4 | 1–6 | Claude Fable 5 (high) | Anthropic | 1510 ±6 | 13415 |
| 5 | 3–7 | Claude Opus 4.7 (high) | Anthropic | 1503 ±5 | 22532 |
| 6 | 3–17 | [Claude Fable 5.1 \(max\)](https://pirouter.ai/zh/models/anthropic/claude-fable-5.1) | Anthropic | 1498 ±11 | 3380 |
| 7 | 4–13 | Claude Opus 5 (high) | Anthropic | 1497 ±6 | 20700 |
| 8 | 6–19 | [GPT-5.6 Sol \(xhigh\)](https://pirouter.ai/zh/models/openai/gpt-5.6-sol) | OpenAI | 1491 ±6 | 12619 |
| 9 | 6–18 | Claude Opus 4.8 (high) | Anthropic | 1490 ±5 | 22488 |
| 10 | 6–23 | Kimi K3 (max) | Moonshot | 1487 ±7 | 9232 |
| 11 | 6–24 | [Gemini 3.8 Flash \(high\)](https://pirouter.ai/zh/models/google/gemini-3.8-flash) | Google | 1487 ±7 | 8909 |
| 12 | 6–29 | Muse Spark 1.3 (max) | Meta | 1486 ±9 | 4263 |
| 13 | 6–29 | [Gemini 3.7 Flash \(high\)](https://pirouter.ai/zh/models/google/gemini-3.7-flash) | Google | 1486 ±8 | 7516 |
| 14 | 7–30 | Claude Opus 4.5 20251101 High 32k | Anthropic | 1483 ±7 | 9867 |
| 15 | 10–31 | Gemini 3.1 Pro Preview | Google | 1479 ±4 | 41331 |
| 16 | 7–41 | [DeepSeek V4.1 Flash \(max\)](https://pirouter.ai/zh/models/deepseek/deepseek-v4.1-flash) | DeepSeek | 1478 ±11 | 2880 |
| 17 | 8–37 | GLM 5.3 (max) | Z.ai | 1478 ±8 | 5877 |
| 18 | 10–35 | [GPT-5.5 \(high\)](https://pirouter.ai/zh/models/openai/gpt-5.5) | OpenAI | 1478 ±5 | 23963 |
| 19 | 12–37 | Claude Opus 4.5 | Anthropic | 1475 ±5 | 21789 |
| 20 | 7–53 | Muse Spark 1.2 (xhigh) | Meta | 1474 ±16 | 1408 |
| 21 | 12–37 | Claude Sonnet 4.6 | Anthropic | 1474 ±5 | 23644 |
| 22 | 12–40 | Muse Spark 1.1 | Meta | 1474 ±6 | 13106 |
| 23 | 12–40 | [Gemini 3.6 Flash \(high\)](https://pirouter.ai/zh/models/google/gemini-3.6-flash) | Google | 1473 ±6 | 12930 |
| 24 | 7–54 | MiMo V2.6 Pro | Xiaomi | 1473 ±16 | 1382 |
| 25 | 9–51 | [GPT-6 Astra \(max\)](https://pirouter.ai/zh/models/openai/gpt-6-astra) | OpenAI | 1473 ±12 | 2708 |
| 26 | 10–43 | [GLM 5.3 Flash](https://pirouter.ai/zh/models/z-ai/glm-5.3-flash) | Z.ai | 1473 ±8 | 7636 |
| 27 | 11–43 | Qwen3.8 (max) | Alibaba | 1473 ±7 | 8169 |
| 28 | 12–42 | Gemini 3 Pro | Google | 1473 ±6 | 11654 |
| 29 | 14–43 | [GPT-5.4 \(high\)](https://pirouter.ai/zh/models/openai/gpt-5.4) | OpenAI | 1471 ±5 | 21840 |
| 30 | 15–46 | [GLM 5.2 \(max\)](https://pirouter.ai/zh/models/z-ai/glm-5.2) | Z.ai | 1470 ±6 | 15700 |
| 31 | 16–47 | MiMo V2.5 Pro | Xiaomi | 1469 ±5 | 23503 |
| 32 | 16–51 | [Claude Sonnet 5 \(high\)](https://pirouter.ai/zh/models/anthropic/claude-sonnet-5) | Anthropic | 1467 ±6 | 15961 |
| 33 | 16–52 | Qwen3.5 Preview (max) | Alibaba | 1467 ±7 | 7395 |
| 34 | 17–52 | [Gemini 3.5 Flash \(medium\)](https://pirouter.ai/zh/models/google/gemini-3.5-flash) | Google | 1466 ±6 | 16877 |
| 35 | 10–67 | Qwen3.7 Preview (max) | Alibaba | 1465 ±16 | 1376 |
| 36 | 16–57 | Muse Spark | Meta | 1465 ±9 | 4729 |
| 37 | 20–52 | Claude Sonnet 4.5 20250929 High 32k | Anthropic | 1464 ±5 | 24596 |
| 38 | 23–54 | Claude Sonnet 4.5 | Anthropic | 1462 ±5 | 24273 |
| 39 | 20–57 | Grok 4.5 | xAI | 1462 ±6 | 14306 |
| 40 | 17–64 | [DeepSeek V4 Pro 0813 \(high\)](https://pirouter.ai/zh/models/deepseek/deepseek-v4-pro) | DeepSeek | 1461 ±10 | 3481 |
| 41 | 20–58 | [GPT-5.6 Terra \(xhigh\)](https://pirouter.ai/zh/models/openai/gpt-5.6-terra) | OpenAI | 1461 ±6 | 13176 |
| 42 | 27–58 | GLM 5.1 | Z.ai | 1460 ±5 | 20257 |
| 43 | 28–60 | Claude Opus 4.1 (thinking) | Anthropic | 1458 ±6 | 13130 |
| 44 | 27–65 | Gemini 3 Flash | Google | 1458 ±7 | 8486 |
| 45 | 29–64 | GPT-5.2 Chat Latest | OpenAI | 1457 ±6 | 11737 |
| 46 | 29–67 | GPT-5.5 Instant | OpenAI | 1456 ±7 | 9680 |
| 47 | 29–65 | Kimi K2.6 | Moonshot | 1456 ±6 | 13174 |
| 48 | 22–77 | Gemma 4 31b | Google | 1453 ±14 | 1801 |
| 49 | 27–74 | [GPT-6 Sol \(max\)](https://pirouter.ai/zh/models/openai/gpt-6-sol) | OpenAI | 1453 ±12 | 2514 |
| 50 | 24–79 | MiMo V2.6 Flash | Xiaomi | 1453 ±14 | 1834 |
| 51 | 34–68 | Ernie 5.1 | Baidu | 1453 ±7 | 12637 |
| 52 | 37–71 | GPT-5.1 (high) | OpenAI | 1450 ±6 | 11148 |
| 53 | 35–71 | Grok 4.6 (high) | xAI | 1450 ±7 | 8225 |
| 54 | 37–71 | Grok 4.20 Beta1 | xAI | 1450 ±7 | 8824 |
| 55 | 29–84 | Qwen3.6 Preview (max) | Alibaba | 1449 ±14 | 1884 |
| 56 | 31–80 | [GPT-6 Luna \(max\)](https://pirouter.ai/zh/models/openai/gpt-6-luna) | OpenAI | 1449 ±12 | 2621 |
| 57 | 41–74 | Qwen3.7 Plus | Alibaba | 1447 ±6 | 14980 |
| 58 | 42–74 | Grok 4.20 Beta 0309 Reasoning | xAI | 1447 ±5 | 22594 |
| 59 | 37–81 | Hy3 | Tencent | 1447 ±10 | 3686 |
| 60 | 41–77 | GLM 5 | Z.ai | 1446 ±7 | 9561 |
| 61 | 42–79 | [GPT-5.6 Luna \(xhigh\)](https://pirouter.ai/zh/models/openai/gpt-5.6-luna) | OpenAI | 1445 ±6 | 13383 |
| 62 | 42–80 | MiMo V2 Pro | Xiaomi | 1444 ±7 | 8445 |
| 63 | 42–80 | Claude Opus 4 (thinking) | Anthropic | 1444 ±7 | 9077 |
| 64 | 46–79 | Grok 4.20 Multi Agent Beta 0309 | xAI | 1444 ±5 | 21911 |
| 65 | 48–79 | Gemini 3 Flash | Google | 1443 ±5 | 29738 |
| 66 | 46–81 | [Gemini 3.5 Flash Lite](https://pirouter.ai/zh/models/google/gemini-3.5-flash-lite) | Google | 1443 ±6 | 12701 |
| 67 | 39–95 | Grok 4.7 (xhigh) | xAI | 1442 ±14 | 1875 |
| 68 | 49–84 | Kimi K2.5 (thinking) | Moonshot | 1439 ±5 | 24154 |
| 69 | 52–85 | Gemini 2.5 Pro | Google | 1438 ±4 | 35029 |
| 70 | 44–99 | Gemma 4 26b A4b | Google | 1437 ±14 | 1737 |
| 71 | 49–95 | GPT-4.5 Preview 2025 02 27 | OpenAI | 1437 ±8 | 5501 |
| 72 | 52–94 | GPT-5.3 Chat Latest | OpenAI | 1436 ±7 | 11195 |
| 73 | 52–94 | Qwen3.6 Plus | Alibaba | 1436 ±6 | 16521 |
| 74 | 55–94 | Dola Seed 2.0 Pro | ByteDance | 1435 ±5 | 26185 |
| 75 | 55–94 | Minimax M3 | MiniMax | 1435 ±5 | 20240 |
| 76 | 57–95 | [GPT-5.4 Mini \(high\)](https://pirouter.ai/zh/models/openai/gpt-5.4-mini) | OpenAI | 1434 ±5 | 21666 |
| 77 | 61–94 | Qwen3.5 397b A17b | Alibaba | 1434 ±5 | 30028 |
| 78 | 57–95 | [DeepSeek V4 Flash 0731 \(high\)](https://pirouter.ai/zh/models/deepseek/deepseek-v4-flash) | DeepSeek | 1434 ±6 | 17993 |
| 79 | 49–104 | Kimi K2.5 Instant | Moonshot | 1432 ±12 | 2335 |
| 80 | 66–98 | Grok 4.1 | xAI | 1431 ±5 | 19477 |
| 81 | 64–98 | MiMo V2.5 | Xiaomi | 1431 ±6 | 16031 |
| 82 | 66–105 | MiMo V2 Omni | Xiaomi | 1428 ±8 | 7037 |
| 83 | 69–103 | GPT-5.2 (high) | OpenAI | 1427 ±6 | 15259 |
| 84 | 69–102 | Chatgpt 4o Latest | OpenAI | 1427 ±4 | 22851 |
| 85 | 68–110 | Qwen3.8 27b | Alibaba | 1426 ±8 | 6306 |
| 86 | 69–105 | Inkling | Thinky | 1426 ±6 | 12442 |
| 87 | 66–112 | GLM 4.7 | Z.ai | 1425 ±10 | 3309 |
| 88 | 69–108 | Ernie 5.0 0110 | Baidu | 1425 ±6 | 11082 |
| 89 | 69–111 | Qwen3 Preview (max) | Alibaba | 1424 ±7 | 7261 |
| 90 | 69–113 | GLM 5v Turbo | Z.ai | 1424 ±10 | 3673 |
| 91 | 69–119 | Mistral Medium 3.5 | Mistral | 1421 ±10 | 3820 |
| 92 | 55–134 | DeepSeek V3.1 Terminus (thinking) | DeepSeek | 1421 ±20 | 850 |
| 93 | 78–113 | DeepSeek V3.2 | DeepSeek | 1420 ±6 | 13522 |
| 94 | 69–123 | Ernie 5.0 Preview 1203 | Baidu | 1420 ±11 | 2681 |
| 95 | 81–116 | Kimi K2 Thinking Turbo | Moonshot | 1418 ±5 | 18691 |
| 96 | 78–127 | DeepSeek V3.1 (thinking) | DeepSeek | 1416 ±11 | 2819 |
| 97 | 81–121 | Longcat Flash Chat 2602 Exp | Meituan | 1416 ±7 | 9619 |
| 98 | 78–129 | DeepSeek V3.2 Exp (thinking) | DeepSeek | 1416 ±12 | 2548 |
| 99 | 81–123 | Claude Sonnet 4 (thinking) | Anthropic | 1415 ±7 | 8466 |
| 100 | 86–119 | Claude Haiku 4.5 | Anthropic | 1415 ±4 | 45457 |

## FAQ

### 指令遵循最好的 AI 模型是哪个？

截至 2026-09-30，指令遵循榜第一为 Google Gemini 4 Argon（high），分数 1529，名次区间 1–4。

### 指令遵循最好的开放权重模型是哪个？

Kimi K3（Moonshot），全榜第 10 名。

### 这个榜单排的是什么？

按格式、长度、步骤等约束作答的排名。

### 多久更新一次？

有新模型发布或排名变化时更新；本榜最近一次更新为 2026-09-30。

数据：[LMArena leaderboard dataset](https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset)，按 [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) 许可使用；仅做筛选与推理档位合并，分数未改动。
查看原榜单: https://arena.ai/leaderboard/text/instruction-following