沙丘社区

垂直行业 · 工业

Leaderboard

工业行业实战能力排行榜

基于 SQBench 工业实战任务结果排序，展示模型在真实业务场景中的综合表现。

排名模型满分 100，越高表示工业实战能力越强未触发风险扣分的任务比例，越高越好触发幻觉相关风险的任务比例，越低越好完成本次评测的模型 API 成本
1GLM-5.2 (Max)智谱43.583.5%10.1%$19.932GLM-5.1智谱41.287.2%5.1%$10.823Claude Opus 4.8 (Max)Anthropic38.382.0%1.5%$133.484MiMo-V2.5小米38.191.0%2.7%$1.045MiniMax-M3MiniMax37.585.2%6.0%$3.906Qwen3.7-Plus阿里巴巴36.592.2%3.6%$4.437Qwen3.7-Max阿里巴巴35.790.3%2.1%$23.048Doubao-Seed-2.1-turbo (High)字节跳动35.587.4%6.0%$5.469Doubao-Seed-2.1-pro (High)字节跳动34.491.3%2.2%$13.5610DeepSeek-V4-Pro (High)DeepSeek34.084.6%6.6%$2.16

Models

工业行业能力表现

展示模型在 SQBench 工业实战任务中的总体表现、完成质量与风险影响。

工业实战能力

基于任务通过情况衡量模型在真实业务任务中的稳定交付表现，越高表示可稳定完成的任务比例越高。

43.5GLM-5.2 (Max)
41.2GLM-5.1
38.3Claude Opus 4.8 (Max)
38.1MiMo-V2.5
37.5MiniMax-M3
36.5Qwen3.7-Plus
35.7Qwen3.7-Max
35.5Doubao-Seed-2.1-turbo (High)
34.4Doubao-Seed-2.1-pro (High)
34.0DeepSeek-V4-Pro (High)
33.5Kimi K2.6
33.2DeepSeek-V4-Flash (High)
32.4DeepSeek-V4-Flash (Max)
32.0MiMo-V2.5-Pro
31.2Kimi K2.7 Code
30.6DeepSeek-V4-Pro (Max)
29.2Qwen3.6-Plus
28.8Hy3
26.8Hy3 Preview
26.4Gemini 3.5 Flash (High)
25.1MiniMax-M2.7
23.7Gemma 4 31B
21.7Grok 4.3 (High)
17.2Step 3.5 Flash (High)
10.7Nemotron 3 Ultra

推理模型以此图标标识

工业实战能力

实战能力采用任务通过口径计算，强调模型在标准评测条件下稳定达标的能力；最终表现分和任务完成分保留连续得分，用于辅助分析任务完成质量与风险影响。

工业实战能力（按开放属性划分）

按开源/闭源展示模型实战能力，用于观察不同开放属性模型的表现分布。

开源模型闭源模型
43.5GLM-5.2 (Max)
41.2GLM-5.1
38.3Claude Opus 4.8 (Max)
38.1MiMo-V2.5
37.5MiniMax-M3
36.5Qwen3.7-Plus
35.7Qwen3.7-Max
35.5Doubao-Seed-2.1-turbo (High)
34.4Doubao-Seed-2.1-pro (High)
34.0DeepSeek-V4-Pro (High)
33.5Kimi K2.6
33.2DeepSeek-V4-Flash (High)
32.4DeepSeek-V4-Flash (Max)
32.0MiMo-V2.5-Pro
31.2Kimi K2.7 Code
30.6DeepSeek-V4-Pro (Max)
29.2Qwen3.6-Plus
28.8Hy3
26.8Hy3 Preview
26.4Gemini 3.5 Flash (High)
25.1MiniMax-M2.7
23.7Gemma 4 31B
21.7Grok 4.3 (High)
17.2Step 3.5 Flash (High)
10.7Nemotron 3 Ultra

推理模型以此图标标识

Capability

工业行业实战能力分析

从数理审计、复杂业务和合规风控三个任务类型，比较模型在工业场景中的实战表现。

数理审计

评估模型核对数据、检查口径一致性和发现计算异常的能力。

40.0Claude Opus 4.8 (Max)
40.0DeepSeek-V4-Flash (Max)
20.0GLM-5.2 (Max)
20.0GLM-5.1
20.0MiMo-V2.5
20.0MiniMax-M3
20.0Qwen3.7-Plus
20.0Qwen3.7-Max
20.0Doubao-Seed-2.1-turbo (High)
20.0Doubao-Seed-2.1-pro (High)
20.0Nemotron 3 Ultra
0.0DeepSeek-V4-Pro (High)
0.0Kimi K2.6
0.0DeepSeek-V4-Flash (High)
0.0MiMo-V2.5-Pro
0.0Kimi K2.7 Code
0.0DeepSeek-V4-Pro (Max)
0.0Qwen3.6-Plus
0.0Hy3
0.0Hy3 Preview
0.0Gemini 3.5 Flash (High)
0.0MiniMax-M2.7
0.0Gemma 4 31B
0.0Grok 4.3 (High)
0.0Step 3.5 Flash (High)

Safety & Reliability

工业行业安全与合规

对比模型在工业任务执行中的可靠性、风险控制和幻觉相关表现。

可靠度

未触发风险扣分的任务比例，越高表示执行过程越稳定。

92.5%Grok 4.3 (High)
92.2%Qwen3.7-Plus
91.3%Doubao-Seed-2.1-pro (High)
91.0%MiMo-V2.5
90.3%Qwen3.7-Max
89.0%Qwen3.6-Plus
88.8%DeepSeek-V4-Flash (Max)
88.1%MiniMax-M2.7
87.4%Doubao-Seed-2.1-turbo (High)
87.2%GLM-5.1
85.3%Gemini 3.5 Flash (High)
85.2%MiniMax-M3
84.6%DeepSeek-V4-Pro (High)
84.3%Gemma 4 31B
83.8%DeepSeek-V4-Pro (Max)
83.5%GLM-5.2 (Max)
83.4%Kimi K2.6
82.1%Kimi K2.7 Code
82.0%Claude Opus 4.8 (Max)
81.6%DeepSeek-V4-Flash (High)
77.8%Hy3
76.4%Hy3 Preview
75.5%MiMo-V2.5-Pro
69.3%Step 3.5 Flash (High)
66.2%Nemotron 3 Ultra

推理模型以此图标标识

Cost Efficiency

工业行业成本效率

展示模型完成 SQBench 工业实战任务所需的评测成本、完成耗时和 Token 消耗，用于辅助判断规模化使用成本。

评测成本

完成本次评测产生的模型 API 成本，越低表示调用成本越低。

输入推理回答
$0.90DeepSeek-V4-Flash (High)$0.90输入$0.45推理$0.23回答$0.22DeepSeek-V4-Flash (High)
$0.90DeepSeek-V4-Flash (Max)$0.90输入$0.39推理$0.30回答$0.21DeepSeek-V4-Flash (Max)
$1.04Hy3$1.04输入$0.55推理$0.00回答$0.49Hy3
$1.04MiMo-V2.5$1.04输入$0.53推理$0.30回答$0.22MiMo-V2.5
$1.28Gemma 4 31B$1.28输入$1.08推理$0.00回答$0.20Gemma 4 31B
$1.36Step 3.5 Flash (High)$1.36输入$0.85推理$0.00回答$0.51Step 3.5 Flash (High)
$2.16DeepSeek-V4-Pro (High)$2.16输入$0.94推理$0.61回答$0.61DeepSeek-V4-Pro (High)
$2.78MiMo-V2.5-Pro$2.78输入$1.48推理$0.67回答$0.64MiMo-V2.5-Pro
$3.15DeepSeek-V4-Pro (Max)$3.15输入$1.72推理$0.81回答$0.63DeepSeek-V4-Pro (Max)
$3.90MiniMax-M3$3.90输入$1.78推理$0.77回答$1.35MiniMax-M3
$3.93Hy3 Preview$3.93输入$3.14推理$0.00回答$0.79Hy3 Preview
$3.94MiniMax-M2.7$3.94输入$2.05推理$0.46回答$1.42MiniMax-M2.7
$4.43Qwen3.7-Plus$4.43输入$1.94推理$1.36回答$1.13Qwen3.7-Plus
$5.46Doubao-Seed-2.1-turbo (High)$5.46输入$2.18推理$1.63回答$1.64Doubao-Seed-2.1-turbo (High)
$8.58Grok 4.3 (High)$8.58输入$5.87推理$2.06回答$0.65Grok 4.3 (High)
$9.98$8.36Nemotron 3 Ultra$9.98输入$8.36推理$0.96回答$0.66Nemotron 3 Ultra
$10.82GLM-5.1$10.82输入$6.27推理$2.57回答$1.98GLM-5.1
$11.04$8.46Qwen3.6-Plus$11.04输入$8.46推理$1.17回答$1.41Qwen3.6-Plus
$11.34Kimi K2.7 Code$11.34输入$5.42推理$3.96回答$1.95Kimi K2.7 Code
$13.56Doubao-Seed-2.1-pro (High)$13.56输入$6.20推理$3.47回答$3.89Doubao-Seed-2.1-pro (High)
$15.35$7.16Kimi K2.6$15.35输入$7.16推理$3.36回答$4.84Kimi K2.6
$19.93$9.75$7.31GLM-5.2 (Max)$19.93输入$9.75推理$7.31回答$2.87GLM-5.2 (Max)
$23.04$15.44Qwen3.7-Max$23.04输入$15.44推理$4.15回答$3.46Qwen3.7-Max
$51.85$29.82$15.42Gemini 3.5 Flash (High)$51.85输入$29.82推理$15.42回答$6.61Gemini 3.5 Flash (High)
$133.48$44.25$9.40$79.83Claude Opus 4.8 (Max)$133.48输入$44.25推理$9.40回答$79.83Claude Opus 4.8 (Max)

推理模型以此图标标识

Value

工业行业性价比

对比模型工业实战能力与可靠度、评测成本、完成耗时和 Token 消耗，辅助判断不同模型的部署价值。

工业实战能力 vs 可靠度

对比模型的任务交付表现与风险稳定性，右上区域更优。

智谱Anthropic小米MiniMax阿里巴巴字节跳动DeepSeek月之暗面腾讯GooglexAI阶跃星辰Nvidia
理想区域工业实战能力可靠度10.020.030.040.070.0%80.0%90.0%GLM-5.2 (Max)GLM-5.2 (Max)可靠度83.5%工业实战能力43.5%Click for more infoGLM-5.1GLM-5.1可靠度87.2%工业实战能力41.2%Click for more infoClaude Opus 4.8 (Max)Claude Opus 4.8 (Max)可靠度82.0%工业实战能力38.3%Click for more infoMiMo-V2.5MiMo-V2.5可靠度91.0%工业实战能力38.1%Click for more infoMiniMax-M3MiniMax-M3可靠度85.2%工业实战能力37.5%Click for more infoQwen3.7-PlusQwen3.7-Plus可靠度92.2%工业实战能力36.5%Click for more infoQwen3.7-MaxQwen3.7-Max可靠度90.3%工业实战能力35.7%Click for more infoDoubao-Seed-2.1-turbo (High)Doubao-Seed-2.1-turbo (High)可靠度87.4%工业实战能力35.5%Click for more infoDoubao-Seed-2.1-pro (High)Doubao-Seed-2.1-pro (High)可靠度91.3%工业实战能力34.4%Click for more infoDeepSeek-V4-Pro (High)DeepSeek-V4-Pro (High)可靠度84.6%工业实战能力34.0%Click for more infoKimi K2.6Kimi K2.6可靠度83.4%工业实战能力33.5%Click for more infoDeepSeek-V4-Flash (High)DeepSeek-V4-Flash (High)可靠度81.6%工业实战能力33.2%Click for more infoDeepSeek-V4-Flash (Max)DeepSeek-V4-Flash (Max)可靠度88.8%工业实战能力32.4%Click for more infoMiMo-V2.5-ProMiMo-V2.5-Pro可靠度75.5%工业实战能力32.0%Click for more infoKimi K2.7 CodeKimi K2.7 Code可靠度82.1%工业实战能力31.2%Click for more infoDeepSeek-V4-Pro (Max)DeepSeek-V4-Pro (Max)可靠度83.8%工业实战能力30.6%Click for more infoQwen3.6-PlusQwen3.6-Plus可靠度89.0%工业实战能力29.2%Click for more infoHy3Hy3可靠度77.8%工业实战能力28.8%Click for more infoHy3 PreviewHy3 Preview可靠度76.4%工业实战能力26.8%Click for more infoGemini 3.5 Flash (High)Gemini 3.5 Flash (High)可靠度85.3%工业实战能力26.4%Click for more infoMiniMax-M2.7MiniMax-M2.7可靠度88.1%工业实战能力25.1%Click for more infoGemma 4 31BGemma 4 31B可靠度84.3%工业实战能力23.7%Click for more infoGrok 4.3 (High)Grok 4.3 (High)可靠度92.5%工业实战能力21.7%Click for more infoStep 3.5 Flash (High)Step 3.5 Flash (High)可靠度69.3%工业实战能力17.2%Click for more infoNemotron 3 UltraNemotron 3 Ultra可靠度66.2%工业实战能力10.7%Click for more info

推理模型以此图标标识