Kimi K2.5 (Non-reasoning)
KimiKimiOpen WeightMIT · Commercial OK
Описание
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.
Дата выхода
2026-01-27
Параметры
1.0T
Длина контекста
262K
Модальности
image, text, video
Радар способностей
32
general
28
coding
79
reasoning
52
scienceоцен.
50
agents
80
multimodal
Science использует прокси на основе рассуждений, когда специализированные научные бенчмарки недоступны.
Рейтинги
| Домен | #Место | Оценка | Источник |
|---|---|---|---|
| Agents & Tools | 34 | 62.0 | LS |
| Code Ranking | 154 | 49.0 | AA |
| General Ranking | 131 | 61.0 | AA |
| Multimodal Ranking | 58 | 71.0 | LS |
| Reasoning | 67 | 57.0 | LS |
| Science | 117 | 58.0 | AA |
Оценки бенчмарков (LLM Stats)
Agents
WideSearch
79.0%Сам.
DeepSearchQA
77.1%Сам.
BrowseComp
74.9%Сам.
PaperBench
63.5%Сам.
Terminal-Bench 2.0
50.8%Сам.
SWE-Bench Pro
50.7%Сам.
CyberGym
41.3%Сам.
Biology
GPQA
87.6%Сам.
SciCode
48.7%Сам.
Code
SWE-Bench Verified
76.8%Сам.
SWE-bench Multilingual
73.0%Сам.
OJBench (C++)
57.4%Сам.
Economics
FinSearchComp T2&T3
67.8%Сам.
Finance
MMLU-Pro
87.1%Сам.
General
LiveCodeBench v6
85.0%Сам.
MMMU-Pro
78.5%Сам.
SimpleVQA
0.71 / 100Сам.
LongBench v2
61.0%Сам.
Healthcare
VideoMMMU
86.6%Сам.
Image To Text
OCRBench
92.3%Сам.
Long Context
LongVideoBench
79.8%Сам.
LVBench
75.9%Сам.
AA-LCR
70.0%Сам.
Math
AIME 2025
96.1%Сам.
HMMT 2025
95.4%Сам.
MathVista-Mini
90.1%Сам.
MathVision
84.2%Сам.
IMO-AnswerBench
81.8%Сам.
Humanity's Last Exam
50.2%Сам.
Multimodal
InfoVQAtest
92.6%Сам.
OmniDocBench 1.5
88.8%Сам.
Video-MME
87.4%Сам.
MMVU
80.4%Сам.
CharXiv-R
77.5%Сам.
MotionBench
70.4%Сам.
WorldVQA
46.3%Сам.
ZEROBench
0.11 / 100Сам.
Reasoning
Seal-0
57.4%Сам.
Индексы оценки AA
Intelligence Index37.3
Coding Index25.8
Tau20.8
Gpqa0.8
Lcr0.6
Ifbench0.4
Scicode0.4
Terminalbench Hard0.2
Hle0.1
Оценки категорий LLM Stats
Finance90
Language90
Legal90
Video80
Vision80
Frontend Development80
Image To Text80
Long Context80
Math80
Multimodal80
Structured Output70
Biology70
Chemistry70
General70
Healthcare70
Physics70
Reasoning70
Search70
Agents60
Code60
Tool Calling50
Safety40
Цены
Цена ввода$0.6 / 1M tokens
Цена вывода$3 / 1M tokens
Смешанная цена (3:1)$1.2 / 1M tokens
Скорость
Токенов/сек45.6 tokens/s
Задержка первого токена1.22s
Время до первого ответа1.22s
Доступные провайдеры
(Внутренние единицы LS)| Провайдер | Цена ввода | Цена вывода |
|---|---|---|
| Moonshot AI | 600K | 3.0M |
| Fireworks | 600K | 3.0M |