DataLearner 标志

DeepSeek-V4-FlashvsGLM 5.1

在 11 个共同 benchmark 中,GLM 5.1 整体领先:DeepSeek-V4-Flash 领先 2 项,GLM 5.1 领先 9 项,持平 0 项,平均分差 -8.46。

DeepSeek-AI
DeepSeek-V4-Flash

DeepSeek-AI · 2026-04-24 · 推理大模型

智谱AI
GLM 5.1

智谱AI · 2026-03-27 · 推理大模型

DeepSeek-V4-Flash2 (18%)(82%)9 GLM 5.1

评测分数

按能力类目分组,每组内按分差大小排列;共 11 项。

Coding and Software Engineer

GLM 5.1 领先 2/3
评测项DeepSeek-V4-FlashGLM 5.1分差
Text Arena (Coding)1,5776 / 35Thinking High (No Tools)1,53413 / 35Normal (No Tools)+42.54
WeirdML v243.7644 / 52Thinking High (With Tools)57.1032 / 52Normal (With Tools)-13.34
SWE-Bench Pro - Public49.1047 / 57Normal (With Tools)58.4017 / 57Thinking (With Tools)-9.30

AI Agent - Tool Usage

胶着 2/2
评测项DeepSeek-V4-FlashGLM 5.1分差
Terminal-Bench 2.182.7016 / 44Max (With Tools)58.7038 / 44Thinking High (With Tools)+24
Terminal Bench 2.049.1036 / 48Normal (With Tools)63.5013 / 48Thinking (With Tools)-14.40

General Knowledge

GLM 5.1 领先 2/2
评测项DeepSeek-V4-FlashGLM 5.1分差
HLE8.10164 / 181Normal (No Tools)52.3021 / 181Thinking (With Tools)-44.20
LiveBench67.2549 / 115Normal (No Tools)70.1837 / 115Normal (No Tools)-2.93

AI Agent - Information Search

GLM 5.1 领先 1/1
评测项DeepSeek-V4-FlashGLM 5.1分差
BrowseComp73.2028 / 54极高强度思考(工具)79.3017 / 54Thinking (With Tools + Internet)-6.10

Math and Reasoning

GLM 5.1 领先 1/1
评测项DeepSeek-V4-FlashGLM 5.1分差
IMO-AnswerBench41.9022 / 23Normal (No Tools)83.8014 / 23Thinking (No Tools)-41.90

常识推理

GLM 5.1 领先 1/1
评测项DeepSeek-V4-FlashGLM 5.1分差
SimpleBench46.3036 / 67Normal (No Tools)58.7022 / 67Normal (No Tools)-12.40

科学与综合推理

GLM 5.1 领先 1/1
评测项DeepSeek-V4-FlashGLM 5.1分差
GPQA Diamond71.20151 / 226Normal (No Tools)86.2078 / 226Thinking (No Tools)-15

规格对比

字段DeepSeek-V4-FlashGLM 5.1
发布机构DeepSeek-AI智谱AI
发布时间2026-04-242026-03-27
模型类型推理大模型推理大模型
架构MoE 架构MoE 架构
参数规模2840亿754亿
上下文长度1M200K
最大输出384K125K

API 调用价格

价格优先使用 DataLearner 配置的 API 记录;缺失项不做推测。

价格项DeepSeek-V4-FlashGLM 5.1
文本输入$0.14 / 1M tokens$1.4 / 1M tokens
文本输出$0.28 / 1M tokens$4.4 / 1M tokens
缓存读取$0.0028 / 1M tokens$4.4 / 1M tokens
缓存写入暂无公开价格$0.26 / 1M tokens

小结

  • GLM 5.1在以下类目领先:Coding and Software Engineer (2/3)、General Knowledge (2/2)、AI Agent - Information Search (1/1)、Math and Reasoning (1/1)、常识推理 (1/1)、科学与综合推理 (1/1)
  • 胶着类目:AI Agent - Tool Usage

11 个共同 benchmark 上,GLM 5.1 平均高出 8.46 分。

单项差距最大的 benchmark:HLE — DeepSeek-V4-Flash 8.10,GLM 5.1 52.30(分差 -44.20)。

本页正文由结构化模型、价格与 benchmark 数据生成,不使用实时 LLM 撰写。