Kimi-K2-Thinking
About Kimi-K2-Thinking
Kimi K2 Thinking is the latest, most capable version of open-source thinking model. Starting with Kimi K2, we built it as a thinking agent that reasons step-by-step while dynamically invoking tools. It sets a new state-of-the-art on Humanity's Last Exam (HLE), BrowseComp, and other benchmarks by dramatically scaling multi-step reasoning depth and maintaining stable tool-use across 200–300 sequential calls. At the same time, K2 Thinking is a native INT4 quantization model with 262k context window, achieving lossless reductions in inference latency and GPU memory usage
Explore how Kimi-K2-Thinking's deep reasoning, stable long-horizon agency, and extensive 256k context window can autonomously solve complex, multi-step challenges across diverse domains.
Autonomous Dev Agent
Orchestrate multi-step coding workflows, from design to deployment, by reasoning through requirements, generating code, and integrating tests with stable tool use.
Use Case Example:
"Autonomously developed a new microservice in Rust, including API design, database schema, and unit tests, by interacting with a Git repository and CI/CD tools over 150 steps."
Legal & Compliance AI
Analyze vast legal documents and regulatory frameworks (256k context) to identify inconsistencies, compliance gaps, and potential risks through multi-step logical deduction.
Use Case Example:
"Reviewed a 1000-page international trade agreement against specific national regulations, flagging 7 critical clauses requiring amendment, a task that typically took a team of lawyers weeks."
Engineering Design Opt.
Validate intricate engineering designs by simulating performance, optimizing parameters, and identifying potential failure points through iterative reasoning and tool interaction.
Use Case Example:
"Optimized the thermal management system for a satellite by iteratively running FEA simulations and adjusting material properties, reducing peak temperature by 10% through an autonomous 200-step process."
Dynamic Market Strategy
Continuously monitor global market data, competitor strategies, and news feeds to synthesize actionable insights and generate adaptive business recommendations.
Use Case Example:
"Provided daily strategic updates for a fintech startup by autonomously browsing financial news, competitor product launches, and social media sentiment, identifying a new market niche and recommending a pivot in product messaging."
Metadata
Specification
State
Deprecated
Architecture
Calibrated
Yes
Mixture of Experts
Yes
Total Parameters
1000B
Activated Parameters
32B
Reasoning
No
Precision
FP8
Context length
262K
Max Tokens
262K
Compare with Other Models
See how this model stacks up against others.
Moonshot AI
chat
Kimi-K2.5
Release on: Jan 30, 2026
Total Context:
262K
Max output:
262K
Input:
$
0.23
/ M Tokens
Output:
$
3.0
/ M Tokens
Moonshot AI
chat
Kimi-K2-Thinking
Release on: Nov 7, 2025
Total Context:
262K
Max output:
262K
Input:
$
0.55
/ M Tokens
Output:
$
2.5
/ M Tokens
Moonshot AI
chat
Kimi-K2-Instruct-0905
Release on: Sep 8, 2025
Total Context:
262K
Max output:
262K
Input:
$
0.4
/ M Tokens
Output:
$
2
/ M Tokens
Moonshot AI
chat
Kimi-K2-Instruct
Release on: Jul 13, 2025
Total Context:
131K
Max output:
131K
Input:
$
0.58
/ M Tokens
Output:
$
2.29
/ M Tokens
Moonshot AI
chat
Kimi-Dev-72B
Release on: Jun 19, 2025
Total Context:
131K
Max output:
131K
Input:
$
0.29
/ M Tokens
Output:
$
1.15
/ M Tokens
