DeepSeek-V4-ProvsKimi K2.6

在 9 个共同 benchmark 中,Kimi K2.6 整体领先:DeepSeek-V4-Pro 领先 1 项,Kimi K2.6 领先 8 项,持平 0 项,平均分差 -19.42。

DeepSeek-AI
DeepSeek-V4-Pro

DeepSeek-AI · 2026-04-24 · 推理大模型

Moonshot AI
Kimi K2.6

Moonshot AI · 2026-04-20 · 推理大模型

DeepSeek-V4-Pro1 (11%)(89%)8 Kimi K2.6

评测分数

按能力类目分组,每组内按分差大小排列;共 9 项。

Coding and Software Engineer

Kimi K2.6 领先 4/4
评测项DeepSeek-V4-ProKimi K2.6分差
LiveCodeBench56.8075 / 120Normal (No Tools)89.607 / 120Thinking (No Tools)-32.80
SWE-bench Multilingual69.8015 / 20Normal (With Tools)76.704 / 20Thinking (With Tools)-6.90
SWE-bench Verified73.6041 / 108Normal (With Tools)80.2013 / 108Thinking (With Tools)-6.60
SWE-Bench Pro - Public52.1028 / 43Normal (With Tools)58.607 / 43Thinking (With Tools)-6.50

General Knowledge

Kimi K2.6 领先 2/2
评测项DeepSeek-V4-ProKimi K2.6分差
HLE7.70141 / 157Normal (No Tools)549 / 157Thinking (With Tools + Internet)-46.30
GPQA Diamond72.90102 / 178Normal (No Tools)90.5016 / 178Thinking (No Tools)-17.60

AI Agent - Information Search

DeepSeek-V4-Pro 领先 1/1
评测项DeepSeek-V4-ProKimi K2.6分差
BrowseComp83.409 / 45极高强度思考(工具)83.2010 / 45Thinking (With Tools + Internet)+0.20

AI Agent - Tool Usage

Kimi K2.6 领先 1/1
评测项DeepSeek-V4-ProKimi K2.6分差
Terminal Bench 2.059.1022 / 46Normal (With Tools)66.7010 / 46Thinking (With Tools)-7.60

Math and Reasoning

Kimi K2.6 领先 1/1
评测项DeepSeek-V4-ProKimi K2.6分差
IMO-AnswerBench35.3019 / 19Normal (No Tools)866 / 19Thinking (No Tools)-50.70

规格对比

字段DeepSeek-V4-ProKimi K2.6
发布机构DeepSeek-AIMoonshot AI
发布时间2026-04-242026-04-20
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1.6万亿1万亿
上下文长度1M256K
最大输出375K暂无数据

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-ProKimi K2.6
文本输入$0.435 / 1M tokens$0.95 / 1M tokens
文本输出$0.87 / 1M tokens$4 / 1M tokens
缓存读取$0.87 / 1M tokens$0.16 / 1M tokens
缓存写入$0.003625 / 1M tokens$0.95 / 1M tokens

小结

  • DeepSeek-V4-Pro在以下类目领先:AI Agent - Information Search (1/1)
  • Kimi K2.6在以下类目领先:Coding and Software Engineer (4/4)、General Knowledge (2/2)、AI Agent - Tool Usage (1/1)、Math and Reasoning (1/1)

9 个共同 benchmark 上,Kimi K2.6 平均高出 19.42 分。

单项差距最大的 benchmark:IMO-AnswerBench — DeepSeek-V4-Pro 35.30,Kimi K2.6 86(分差 -50.70)。

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。