跳转到主要内容

Kimi K2.6 (Non-reasoning)

KimiKimiOpen WeightModified MIT License

描述

Kimi K2.6 is Moonshot AI's open-source, native multimodal agentic model focused on state-of-the-art coding, long-horizon execution, and agent swarm capabilities. It scales horizontally to 300 sub-agents executing 4,000 coordinated steps, dynamically decomposing tasks into parallel, domain-specialized subtasks. K2.6 unifies text, image, and video input with thinking and non-thinking modes, supports a 256K context, and powers proactive 24/7 background agents that manage schedules, execute code, and orchestrate cross-platform operations without human oversight.

发布日期
2026-04-20
参数规模
1.0T
上下文长度
262K
支持模态
image, text, video

能力雷达图

38
general
39
coding
79
reasoning
53
science估算
60
agents
80
multimodal

Science 在缺少专门科学评测时使用推理能力代理估算。

排行榜排名

领域#排名分数来源
智能体与工具20
68.0
LS
代码能力榜69
67.0
AA
通用能力榜88
70.0
AA
多模态榜22
84.0
LS
推理能力62
61.0
LS
科学能力90
63.0
AA

基准测试分数 (LLM Stats)

Agents

BrowseComp86.3%自报
DeepSearchQA83.0%自报
Claw-Eval80.9%自报
WideSearch80.8%自报
OSWorld-Verified73.1%自报
Terminal-Bench 2.066.7%自报
SWE-Bench Pro58.6%自报
MCP-Mark55.9%自报
Toolathlon50.0%自报
APEX-Agents27.9%自报

Biology

GPQA90.5%自报
SciCode52.2%自报

Code

SWE-Bench Verified80.2%自报
SWE-bench Multilingual76.7%自报

General

LiveCodeBench v689.6%自报
MMMU-Pro80.1%自报

Math

AIME 202696.4%自报
MathVision93.2%自报
HMMT Feb 2692.7%自报
IMO-AnswerBench86.0%自报
Humanity's Last Exam36.4%自报

Multimodal

V*96.9%自报
CharXiv-R86.7%自报
BabyVision68.5%自报

Reasoning

OJBench60.6%自报

AA 评测指数

Intelligence Index
43.0
Coding Index
38.4
Tau2
0.9
Gpqa
0.8
Lcr
0.6
Ifbench
0.4
Scicode
0.4
Terminalbench Hard
0.4
Hle
0.2

LLM Stats 分类评分

Vision
80
Coding
80
Frontend Development
80
General
80
Math
80
Multimodal
80
Reasoning
80
Search
80
Agents
70
Biology
70
Chemistry
70
Code
70
Physics
70
Tool Calling
60

定价

输入价格$0.95 / 1M tokens
输出价格$4 / 1M tokens
混合价格(3:1)$1.712 / 1M tokens

速度

Tokens/秒28.5 tokens/s
首Token延迟1.71s
首回答延迟1.71s

可用提供商

(LS 内部计价单位)
提供商输入价格输出价格
Novita950K4.0M
Moonshot AI950K4.0M

外部链接