Kimi K2.5 (Non-reasoning)
KimiKimi开源权重MIT · 商用许可
描述
Kimi K2.5 is Moonshot AI's flagship agentic model and a new SOTA open model. It unifies vision and text, thinking and non-thinking modes, and single-agent and multi-agent execution into one model. Built with Full-Parameter RL tuning, it achieves state-of-the-art performance across agents, coding, image, and video benchmarks.
发布日期
2026-01-27
参数规模
1.0T
上下文长度
262K
支持模态
image, text, video
能力雷达图
70
general
50
coding
80
reasoning
60
science估算
50
agents
80
multimodal
缺少专门科学评测时,Science 由 LLM Stats 科学得分或推理能力估算。
排行榜排名
| 领域 | #排名 | 分数 | 来源 |
|---|---|---|---|
| 多模态榜 | 35 | 60.0 | LS |
基准测试分数 (LLM Stats)
(LLM Stats (zeroeval))Code
PaperBench
63.5%自报
CyberGym
41.3%自报
FrontierSWE
26.0%
Finance
FinSearchComp T2&T3
67.8%自报
Language
MMLU-Pro
87.1%自报
Long Context
LongBench v2
61.0%自报
Math
AIME 2025
96.1%自报
HMMT 2025
95.4%自报
MathVista-Mini
90.1%自报
MathVision
84.2%自报
IMO-AnswerBench
81.8%自报
LiveBench
69.1%
Multimodal
Video-MME
87.4%自报
VideoMMMU
86.6%自报
Reasoning
GPQANYU + Cohere + Anthropic (2023)
87.6%自报
LiveCodeBench v6
85.0%自报
CharXiv-R
77.5%自报
SWE-Bench Verified
76.8%自报
BrowseCompOpenAI (2025)
74.9%自报
SWE-bench Multilingual
73.0%自报
AA-LCR
70.0%自报
OJBench (C++)
57.4%自报
Seal-0
57.4%自报
Terminal-Bench 2.0Stanford × Laude Institute (2026)
50.8%自报
SWE-Bench ProPrinceton NLP (2024)
50.7%自报
Humanity's Last Exam
50.2%自报
SciCode
48.7%自报
Search
WideSearch
79.0%自报
DeepSearchQA
77.1%自报
Vision
InfoVQAtest
92.6%自报
OCRBench
92.3%自报
OmniDocBench 1.5
88.8%自报
MMVU
80.4%自报
LongVideoBench
79.8%自报
MMMU-Pro
78.5%自报
LVBench
75.9%自报
SimpleVQA
0.71 / 100自报
MotionBench
70.4%自报
WorldVQA
46.3%自报
ZEROBench
0.11 / 100自报
AA 评测指数
(Artificial Analysis)暂无 AA 评测数据
LLM Stats 分类评分
(LLM Stats (zeroeval))Language90
Legal90
Finance90
Image To Text80
Long Context80
Math80
Frontend Development80
Video80
Multimodal70
Physics70
Reasoning70
Search70
Structured Output70
General70
Healthcare70
Biology70
Chemistry70
Vision70
Agents60
Code50
Tool Calling50
Safety40
定价
输入价格$0.6 / 1M tokens
输出价格$3 / 1M tokens
混合价格(3:1)$1 / 1M tokens
缓存读取价格$0.15 / 1M tokens
速度
Tokens/秒54.0
首Token延迟2.26s
首回答延迟2.26s
供应商价格排行
供应商价格排行
11 个供应商
最便宜: OpenRouter最贵: HPC-AI
供应商输入输出
1OpenRouter最便宜
$0.5
$2.5
2TensorX
$0.5
$2.8
3IteraCompute
$0.54
$2.78
4NanoGPT
$0.557
$2.832
5Qiniu
$0.57971
$3.04348
6Kimi主要
$0.6
$3
7Jiekou.AI
$0.6
$3
8Novita AI
$0.6
$3
9Kilo Gateway
$0.6
$3
10Vercel AI Gateway
$0.6
$3
11HPC-AI
$0.6
$3
比较该模型在不同 API 供应商之间的定价。