沙丘社区

垂直行业 · 医疗

Leaderboard

医疗行业实战能力排行榜

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

排名模型满分 100，越高表示医疗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
1Claude Opus 4.8 (Max)Anthropic46.378.0%13.5%$134.932GLM-5.2 (Max)智谱39.591.5%6.1%$20.223Kimi K2.7 Code月之暗面39.278.1%9.6%$11.214DeepSeek-V4-Pro (Max)DeepSeek38.679.8%5.6%$3.115Kimi K2.6月之暗面37.579.4%5.6%$14.876MiniMax-M3MiniMax37.585.2%6.0%$3.957MiMo-V2.5-Pro小米36.075.5%7.6%$2.738Doubao-Seed-2.1-turbo (High)字节跳动35.571.4%18.0%$5.249DeepSeek-V4-Pro (High)DeepSeek34.084.6%6.6%$2.1510DeepSeek-V4-Flash (High)DeepSeek33.277.6%1.6%$0.88

Models

医疗行业能力表现

展示模型在 SQBench 医疗实战任务中的总体表现、完成质量与风险影响。

医疗实战能力

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。

46.3Claude Opus 4.8 (Max)
39.5GLM-5.2 (Max)
39.2Kimi K2.7 Code
38.6DeepSeek-V4-Pro (Max)
37.5Kimi K2.6
37.5MiniMax-M3
36.0MiMo-V2.5-Pro
35.5Doubao-Seed-2.1-turbo (High)
34.0DeepSeek-V4-Pro (High)
33.2DeepSeek-V4-Flash (High)
33.2GLM-5.1
33.2Qwen3.6-Plus
32.8Hy3
32.5Qwen3.7-Plus
31.7Qwen3.7-Max
30.8Hy3 Preview
30.4Doubao-Seed-2.1-pro (High)
30.4Gemini 3.5 Flash (High)
30.1MiMo-V2.5
29.1MiniMax-M2.7
28.4DeepSeek-V4-Flash (Max)
25.7Grok 4.3 (High)
23.7Gemma 4 31B
21.2Step 3.5 Flash (High)
16.7Step 3.7 Flash (High)

推理模型以此图标标识

医疗实战能力

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。

医疗实战能力（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。

闭源模型开源模型
46.3Claude Opus 4.8 (Max)
39.5GLM-5.2 (Max)
39.2Kimi K2.7 Code
38.6DeepSeek-V4-Pro (Max)
37.5Kimi K2.6
37.5MiniMax-M3
36.0MiMo-V2.5-Pro
35.5Doubao-Seed-2.1-turbo (High)
34.0DeepSeek-V4-Pro (High)
33.2DeepSeek-V4-Flash (High)
33.2GLM-5.1
33.2Qwen3.6-Plus
32.8Hy3
32.5Qwen3.7-Plus
31.7Qwen3.7-Max
30.8Hy3 Preview
30.4Doubao-Seed-2.1-pro (High)
30.4Gemini 3.5 Flash (High)
30.1MiMo-V2.5
29.1MiniMax-M2.7
28.4DeepSeek-V4-Flash (Max)
25.7Grok 4.3 (High)
23.7Gemma 4 31B
21.2Step 3.5 Flash (High)
16.7Step 3.7 Flash (High)

推理模型以此图标标识

Capability

医疗行业实战能力分析

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

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。

60.0Claude Opus 4.8 (Max)
40.0GLM-5.2 (Max)
40.0Kimi K2.7 Code
40.0DeepSeek-V4-Pro (Max)
40.0Kimi K2.6
40.0MiMo-V2.5-Pro
20.0MiniMax-M3
20.0Doubao-Seed-2.1-turbo (High)
20.0DeepSeek-V4-Pro (High)
20.0DeepSeek-V4-Flash (High)
20.0GLM-5.1
20.0Qwen3.6-Plus
20.0Hy3
20.0Qwen3.7-Plus
20.0Qwen3.7-Max
20.0Hy3 Preview
20.0Doubao-Seed-2.1-pro (High)
20.0Gemini 3.5 Flash (High)
20.0MiMo-V2.5
20.0MiniMax-M2.7
20.0DeepSeek-V4-Flash (Max)
20.0Grok 4.3 (High)
20.0Gemma 4 31B
20.0Step 3.5 Flash (High)
0.0Step 3.7 Flash (High)

Safety & Reliability

医疗行业安全与合规

对比模型在医疗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。

91.5%GLM-5.2 (Max)
85.2%MiniMax-M3
84.6%DeepSeek-V4-Pro (High)
82.3%Qwen3.7-Max
80.2%Qwen3.7-Plus
79.8%DeepSeek-V4-Pro (Max)
79.4%Kimi K2.6
78.1%Kimi K2.7 Code
78.0%Claude Opus 4.8 (Max)
77.8%Hy3
77.6%DeepSeek-V4-Flash (High)
77.0%Qwen3.6-Plus
76.5%Grok 4.3 (High)
75.5%MiMo-V2.5-Pro
75.0%MiMo-V2.5
73.3%Step 3.5 Flash (High)
72.4%Hy3 Preview
72.1%MiniMax-M2.7
71.4%Doubao-Seed-2.1-turbo (High)
71.3%Doubao-Seed-2.1-pro (High)
71.2%GLM-5.1
69.3%Gemini 3.5 Flash (High)
68.8%DeepSeek-V4-Flash (Max)
67.4%Step 3.7 Flash (High)
64.3%Gemma 4 31B

推理模型以此图标标识

Cost Efficiency

医疗行业成本效率

展示模型完成 SQBench 医疗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
$0.88DeepSeek-V4-Flash (Max)$0.88输入$0.38推理$0.31回答$0.19DeepSeek-V4-Flash (Max)
$0.88DeepSeek-V4-Flash (High)$0.88输入$0.45推理$0.23回答$0.21DeepSeek-V4-Flash (High)
$1.02MiMo-V2.5$1.02输入$0.52推理$0.29回答$0.21MiMo-V2.5
$1.04Hy3$1.04输入$0.55推理$0.00回答$0.49Hy3
$1.26Gemma 4 31B$1.26输入$1.07推理$0.00回答$0.19Gemma 4 31B
$1.31Step 3.5 Flash (High)$1.31输入$0.81推理$0.00回答$0.50Step 3.5 Flash (High)
$2.15DeepSeek-V4-Pro (High)$2.15输入$0.94推理$0.61回答$0.60DeepSeek-V4-Pro (High)
$2.73MiMo-V2.5-Pro$2.73输入$1.45推理$0.66回答$0.62MiMo-V2.5-Pro
$2.95Step 3.7 Flash (High)$2.95输入$1.00推理$0.00回答$1.95Step 3.7 Flash (High)
$3.11DeepSeek-V4-Pro (Max)$3.11输入$1.72推理$0.80回答$0.60DeepSeek-V4-Pro (Max)
$3.87Hy3 Preview$3.87输入$3.10推理$0.00回答$0.76Hy3 Preview
$3.95MiniMax-M3$3.95输入$1.78推理$0.79回答$1.39MiniMax-M3
$3.97MiniMax-M2.7$3.97输入$2.08推理$0.48回答$1.42MiniMax-M2.7
$4.47Qwen3.7-Plus$4.47输入$1.94推理$1.43回答$1.09Qwen3.7-Plus
$5.24Doubao-Seed-2.1-turbo (High)$5.24输入$2.10推理$1.61回答$1.53Doubao-Seed-2.1-turbo (High)
$8.39Grok 4.3 (High)$8.39输入$5.72推理$2.06回答$0.61Grok 4.3 (High)
$10.73GLM-5.1$10.73输入$6.22推理$2.62回答$1.89GLM-5.1
$10.80$8.30Qwen3.6-Plus$10.80输入$8.30推理$1.14回答$1.35Qwen3.6-Plus
$11.21Kimi K2.7 Code$11.21输入$5.39推理$3.92回答$1.91Kimi K2.7 Code
$14.72Doubao-Seed-2.1-pro (High)$14.72输入$6.33推理$3.61回答$4.78Doubao-Seed-2.1-pro (High)
$14.87Kimi K2.6$14.87输入$7.03推理$3.24回答$4.59Kimi K2.6
$20.22$9.86$7.49GLM-5.2 (Max)$20.22输入$9.86推理$7.49回答$2.87GLM-5.2 (Max)
$23.12$15.48Qwen3.7-Max$23.12输入$15.48推理$4.21回答$3.43Qwen3.7-Max
$49.83$28.57$14.81Gemini 3.5 Flash (High)$49.83输入$28.57推理$14.81回答$6.44Gemini 3.5 Flash (High)
$134.93$47.76$9.21$77.96Claude Opus 4.8 (Max)$134.93输入$47.76推理$9.21回答$77.96Claude Opus 4.8 (Max)

推理模型以此图标标识

Value

医疗行业性价比

对比模型医疗实战能力与可靠度、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。

医疗实战能力 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。

Anthropic智谱月之暗面DeepSeekMiniMax小米字节跳动阿里巴巴腾讯GooglexAI阶跃星辰
理想区域医疗实战能力可靠度20.030.040.070.0%80.0%90.0%Claude Opus 4.8 (Max)Claude Opus 4.8 (Max)可靠度78.0%医疗实战能力46.3%Click for more infoGLM-5.2 (Max)GLM-5.2 (Max)可靠度91.5%医疗实战能力39.5%Click for more infoKimi K2.7 CodeKimi K2.7 Code可靠度78.1%医疗实战能力39.2%Click for more infoDeepSeek-V4-Pro (Max)DeepSeek-V4-Pro (Max)可靠度79.8%医疗实战能力38.6%Click for more infoKimi K2.6Kimi K2.6可靠度79.4%医疗实战能力37.5%Click for more infoMiniMax-M3MiniMax-M3可靠度85.2%医疗实战能力37.5%Click for more infoMiMo-V2.5-ProMiMo-V2.5-Pro可靠度75.5%医疗实战能力36.0%Click for more infoDoubao-Seed-2.1-turbo (High)Doubao-Seed-2.1-turbo (High)可靠度71.4%医疗实战能力35.5%Click for more infoDeepSeek-V4-Pro (High)DeepSeek-V4-Pro (High)可靠度84.6%医疗实战能力34.0%Click for more infoDeepSeek-V4-Flash (High)DeepSeek-V4-Flash (High)可靠度77.6%医疗实战能力33.2%Click for more infoGLM-5.1GLM-5.1可靠度71.2%医疗实战能力33.2%Click for more infoQwen3.6-PlusQwen3.6-Plus可靠度77.0%医疗实战能力33.2%Click for more infoHy3Hy3可靠度77.8%医疗实战能力32.8%Click for more infoQwen3.7-PlusQwen3.7-Plus可靠度80.2%医疗实战能力32.5%Click for more infoQwen3.7-MaxQwen3.7-Max可靠度82.3%医疗实战能力31.7%Click for more infoHy3 PreviewHy3 Preview可靠度72.4%医疗实战能力30.8%Click for more infoDoubao-Seed-2.1-pro (High)Doubao-Seed-2.1-pro (High)可靠度71.3%医疗实战能力30.4%Click for more infoGemini 3.5 Flash (High)Gemini 3.5 Flash (High)可靠度69.3%医疗实战能力30.4%Click for more infoMiMo-V2.5MiMo-V2.5可靠度75.0%医疗实战能力30.1%Click for more infoMiniMax-M2.7MiniMax-M2.7可靠度72.1%医疗实战能力29.1%Click for more infoDeepSeek-V4-Flash (Max)DeepSeek-V4-Flash (Max)可靠度68.8%医疗实战能力28.4%Click for more infoGrok 4.3 (High)Grok 4.3 (High)可靠度76.5%医疗实战能力25.7%Click for more infoGemma 4 31BGemma 4 31B可靠度64.3%医疗实战能力23.7%Click for more infoStep 3.5 Flash (High)Step 3.5 Flash (High)可靠度73.3%医疗实战能力21.2%Click for more infoStep 3.7 Flash (High)Step 3.7 Flash (High)可靠度67.4%医疗实战能力16.7%Click for more info

推理模型以此图标标识