Passer au contenu principal

Kimi K2.5 (Non-reasoning)

KimiKimiOpen WeightMIT · Usage Commercial

Description

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.

Date de sortie
2026-01-27
Paramètres
1.0T
Longueur du contexte
262K
Modalités
image, text, video

Radar de capacités

26
general
40
coding
79
reasoning
52
science
50
agents
80
multimodal

Classements

Domaine#RangScoreSource
Capacité agentique130
33.0
LS
Classement codage205
54.0
AA
Classement général208
55.0
AA
Classement multimodal27
57.0
LS
Science194
56.0
AA

Scores de benchmarks (LLM Stats)

(LLM Stats (zeroeval))

Agents

WideSearch79.0%Aut.
DeepSearchQA77.1%Aut.
BrowseCompOpenAI (2025)74.9%Aut.
PaperBench63.5%Aut.
Terminal-Bench 2.0Stanford × Laude Institute (2026)50.8%Aut.
SWE-Bench ProPrinceton NLP (2024)50.7%Aut.
CyberGym41.3%Aut.
FrontierSWE26.0%

Biology

GPQANYU + Cohere + Anthropic (2023)87.6%Aut.
SciCode48.7%Aut.

Code

SWE-Bench Verified76.8%Aut.
SWE-bench Multilingual73.0%Aut.
OJBench (C++)57.4%Aut.

Economics

FinSearchComp T2&T367.8%Aut.

Finance

MMLU-Pro87.1%Aut.

General

LiveCodeBench v685.0%Aut.
MMMU-Pro78.5%Aut.
SimpleVQA0.71 / 100Aut.
LiveBench69.1%
LongBench v261.0%Aut.

Healthcare

VideoMMMU86.6%Aut.

Image To Text

OCRBench92.3%Aut.

Long Context

LongVideoBench79.8%Aut.
LVBench75.9%Aut.
AA-LCR70.0%Aut.

Math

AIME 202596.1%Aut.
HMMT 202595.4%Aut.
MathVista-Mini90.1%Aut.
MathVision84.2%Aut.
IMO-AnswerBench81.8%Aut.
Humanity's Last Exam50.2%Aut.

Multimodal

InfoVQAtest92.6%Aut.
OmniDocBench 1.588.8%Aut.
Video-MME87.4%Aut.
MMVU80.4%Aut.
CharXiv-R77.5%Aut.
MotionBench70.4%Aut.
WorldVQA46.3%Aut.
ZEROBench0.11 / 100Aut.

Reasoning

Seal-057.4%Aut.

Indices d'évaluation AA

(Artificial Analysis)
Intelligence Index(Artificial Analysis)
30.1
Tau2(Sierra + U Toronto + Vector Institute (2025))
0.8
Gpqa(NYU + Cohere + Anthropic (2023))
0.8
Lcr(Artificial Analysis)
0.7
Ifbench(Google Research (2023))
0.4
Scicode(UIUC + Argonne National Lab (2024))
0.4
Terminalbench Hard(Stanford × Laude Institute (2026))
0.2
Hle(Center for AI Safety + Scale AI (2025))
0.1

Scores par catégorie LLM Stats

(LLM Stats (zeroeval))
Legal
90
Language
90
Finance
90
Long Context
80
Math
80
Image To Text
80
Frontend Development
80
Video
80
Multimodal
70
Physics
70
Reasoning
70
Search
70
Structured Output
70
General
70
Healthcare
70
Biology
70
Chemistry
70
Vision
70
Agents
60
Code
50
Tool Calling
50
Safety
40

Tarification

Prix d'entrée$0.6 / 1M tokens
Prix de sortie$3 / 1M tokens
Prix mixte (3:1)$1.2 / 1M tokens
Prix de lecture cache$0.095 / 1M tokens

Vitesse

Tokens/sec0.0
Délai du premier token0.00s
Temps de réponse0.00s

Classement des Prix par Fournisseur

Classement des Prix par Fournisseur

11 fournisseurs

Moins cher: NanoGPTPlus cher: Ofox
FournisseurEntréeSortie
1NanoGPTMoins cher
$0.3
$1.9
2TensorX
$0.5
$2.8
3OpenRouter
$0.57
$2.85
4ZenMux
$0.58
$3.02
5KimiPRINCIPAL
$0.6
$3
6Jiekou.AI
$0.6
$3
7NovitaAI
$0.6
$3
8Kilo Gateway
$0.6
$3
9Vercel AI Gateway
$0.6
$3
10HPC-AI
$0.6
$3
11Ofox
$0.6
$3

Comparer les prix entre différents fournisseurs API pour ce modèle.

Sources externes