Kimi K2.7 Code
KimiKimiOpen WeightModified MIT License
Description
Kimi K2.7 Code is Moonshot AI's coding-focused agentic model built on Kimi K2.6. It improves real-world long-horizon coding task completion, instruction following, and token efficiency while reducing thinking-token usage by approximately 30% versus Kimi K2.6. The model uses a native multimodal Mixture-of-Experts architecture with 1T total parameters, 32B activated parameters, a 256K-token context window, image and video input support, and forced thinking with preserve_thinking enabled for multi-turn coding agent workflows.
Date de sortie
2026-06-12
Paramètres
1.0T
Longueur du contexte
262K
Modalités
image, text, video
Radar de capacités
35
general
59
coding
90
reasoning
64
science
80
agents
90
multimodal
Classements
| Domaine | #Rang | Score | Source |
|---|---|---|---|
| Capacité agentique | 16 | 64.0 | LS |
| Classement codage | 87 | 77.0 | AA |
| Classement général | 96 | 82.0 | AA |
| Raisonnement mathématique | 51 | 80.0 | LB |
| Science | 76 | 72.0 | AA |
Scores de benchmarks (LLM Stats)
(LLM Stats (zeroeval))Agents
MCP-Mark
81.1%Aut.
Kimi Code Bench v2
62.0%Aut.
Program Bench
53.6%Aut.
Kimi Claw 24/7 Bench
46.9%Aut.
MLS-Bench Lite
35.1%Aut.
Code
DeepSWE 1.1
31.0%
Math
LiveBench
71.9%
Reasoning
MCP Atlas
76.0%Aut.
FrontierCode 1.1
30.1%
Indices d'évaluation AA
(Artificial Analysis)Tau2(Sierra + U Toronto + Vector Institute (2025))90.1
Gpqa(NYU + Cohere + Anthropic (2023))89.6
Lcr(Artificial Analysis)79.3
Terminalbench V2 167.4
Ifbench(Google Research (2023))63.1
Coding Index(Artificial Analysis)60.8
Scicode(UIUC + Argonne National Lab (2024))47.8
Terminalbench Hard(Stanford × Laude Institute (2026))44.7
Hle(Center for AI Safety + Scale AI (2025))35.0
Tau Banking20.2
Terminalbench V4 01.0
Scores par catégorie LLM Stats
(LLM Stats (zeroeval))Tool Calling80
Math70
Reasoning50
General50
Agents50
Code50
Tarification
Prix d'entrée$0.95 / 1M tokens
Prix de sortie$4 / 1M tokens
Prix mixte (3:1)$1.712 / 1M tokens
Prix de lecture cache$0.15 / 1M tokens
Vitesse
Tokens/sec60.4
Délai du premier token1.06s
Temps de réponse37.95s
Classement des Prix par Fournisseur
Classement des Prix par Fournisseur
34 fournisseurs
Moins cher: DeepInfraPlus cher: TensorX
FournisseurEntréeSortie
1DeepInfraMoins cher
$0
$0
2routing.run
$0.275
$1.1
3OpenRouter
$0.56
$3.1
4Kilo Gateway
$0.56
$3.1
5Kenari
$0.57143
$2.85714
6NanoGPT
$0.6
$3
7Melious
$0.61325
$3.28925
8Cortecs
$0.656
$3.3
9IteraCompute
$0.69
$3.29
10DevPass (LLM Gateway)
$0.737
$3.08
11Venice AI
$0.75
$3.5
12GreenPT
$0.86356
$4.31778
13KimiPRINCIPAL
$0.95
$4
14302.AI
$0.95
$4
15OpenCode Go
$0.95
$4
16ZenMux
$0.95
$4
17Novita AI
$0.95
$4
18GitHub Copilot
$0.95
$4
19OpenCode
$0.95
$4
20Ollama Cloud
$0.95
$4
21AIHubMix
$0.95
$3.9995
22Azure Cognitive Services
$0.95
$4
23Requesty
$0.95
$4
24Vercel AI Gateway
$0.95
$4
25Azure
$0.95
$4
26Moonshot AI
$0.95
$4
27Neuralwatt
$0.95
$4
28HPC-AI
$0.95
$4
29EmpirioLabs AI
$0.95
$4
30Ofox
$0.95
$4
31Moonshot AI (China)
$0.9711
$4.0337
32OpenReason
$1.0022
$4.22
33Charm Hyper
$1.03436
$4.3552
34TensorX
$1.25
$4.5
Comparer les prix entre différents fournisseurs API pour ce modèle.