Model

Kimi K2-Instruct-0905

Moonshot AI
开源 MIT

Kimi K2-Instruct-0905 is the latest, most capable version of Kimi K2, achieving state-of-the-art performance in frontier knowledge, math, and coding among non-thinking models. This Mixture-of-Experts model features 32 billion activated parameters and 1 trillion total parameters, meticulously optimized for agentic tasks. Key features include enhanced agentic coding intelligence, extended context length to 256K tokens, and a hybrid architecture trained with MuonClip optimizer on 15.5T tokens. The model achieves 65.8% on SWE-bench Verified (single attempt), 47.3% on SWE-bench Multilingual, and excels at tool use with 70.6% on Tau2-retail. It is a reflex-grade model without long thinking, designed to act and execute complex tasks seamlessly.

发布日期2025年9月5日
参数规模1.0T
上下文长度
许可证MIT
知识截止

Benchmarks

评测成绩

评测基准 类别 分数 来源
Math 500 math reasoning 97.4 来源
Mmlu Redux language reasoning math general 92.7 来源
Ifeval general 89.8 来源
Mmlu general reasoning language math 89.5 来源
Autologi reasoning 89.5 来源
Zebralogic reasoning 89.0 来源
Multiple general language 85.7 来源
Mmlu Pro language reasoning math general 81.1 来源
Acebench general reasoning 76.5 来源
Livebench math reasoning general 76.4 来源
Gpqa reasoning general 75.1 来源
Cnmo 2024 math 74.3 来源
Tau2 Retail communication reasoning 70.6 来源
Aime 2024 math reasoning 69.6 来源
Swe Bench Verified reasoning frontend_development code 65.8 来源
Tau2 Telecom communication reasoning 65.8 来源
Polymath En math reasoning 65.1 来源
Aider Polyglot general code 60.0 来源
Supergpqa reasoning general math legal healthcare finance chemistry economics physics 57.2 来源
Tau2 Airline reasoning communication 56.5 来源
Multichallenge communication reasoning 54.1 来源
Livecodebench reasoning general code 53.7 来源
Aime 2025 math reasoning 49.5 来源
Swe Bench Multilingual reasoning code 47.3 来源
Hmmt 2025 math 38.8 来源
Simpleqa general reasoning 31.0 来源
Ojbench reasoning 27.1 来源
Terminal Bench reasoning code 25.0 来源
Hle reasoning math 4.7 来源

Pricing

API 价格对比

暂无 API 价格。