Kimi K2 Instruct

Preview

moonshotai/kimi-k2-instruct

Try it in Playground

TOKEN SPEED

~250 TPS

INPUT

Text

OUTPUT

Text

CAPABILITIES

Tool Use, JSON Mode

Moonshot AI

model card

Kimi K2 is Moonshot AI's state-of-the-art Mixture-of-Experts (MoE) language model with 1 trillion total parameters and 32 billion activated parameters. Designed for agentic intelligence, it excels at tool use, coding, and autonomous problem-solving across diverse domains.

License: Please note this model is licensed under a modified MIT license that can be found in the model repository.

PRICING

Input

$1.00

1.0M / $1

Output

$3.00

333,333 / $1

LIMITS

CONTEXT WINDOW

131,072

MAX OUTPUT TOKENS

16,384

Key Technical Specifications

Model Architecture

Built on a Mixture-of-Experts (MoE) architecture with 1 trillion total parameters and 32 billion activated parameters. Features 384 experts with 8 experts selected per token, optimized for efficient inference while maintaining high performance. Trained with the innovative Muon optimizer to achieve zero training instability.

Performance Metrics

The Kimi-K2-Instruct model demonstrates exceptional performance across coding, math, and reasoning benchmarks:

LiveCodeBench: 53.7% Pass@1 (top-tier coding performance)
SWE-bench Verified: 65.8% single-attempt accuracy
MMLU (Massive Multitask Language Understanding): 89.5% exact match
Tau2 retail tasks: 70.6% Avg@4

Use Cases

Agentic AI and Tool Use

Leverage the model's advanced tool calling capabilities for building autonomous agents that can interact with external systems and APIs.

Advanced Code Generation

Utilize the model's top-tier performance in coding tasks, from simple scripting to complex software development and debugging.

Complex Problem Solving

Deploy for multi-step reasoning tasks, mathematical problem-solving, and analytical workflows requiring deep understanding.

Multilingual Applications

Take advantage of strong multilingual capabilities for global applications and cross-language understanding tasks.

Best Practices

Provide clear, detailed tool and function definitions with explicit parameters, expected outputs, and constraints for optimal tool use performance.
Structure complex tasks into clear steps to leverage the model's agentic reasoning capabilities effectively.
Use the full 128K context window for complex, multi-step workflows and comprehensive documentation analysis.
Leverage the model's multilingual capabilities by clearly specifying the target language and cultural context when needed.

Get Started with Kimi K2

Experience moonshotai/kimi-k2-instruct on Groq:

shell

pip install groq

Python

from groq import Groq
client = Groq()
completion = client.chat.completions.create(
    model="moonshotai/kimi-k2-instruct",
    messages=[
        {
            "role": "user",
            "content": "Explain why fast inference is critical for reasoning models"
        }
    ]
)
print(completion.choices[0].message.content)