Kimi K2 0905

Kimi K2 0905. Kimi-k2 is a Mixture-of-Experts (MoE) foundation model with exceptional coding and agent capabilities, featuring 1 trillion total parameters and 32 billion activated parameters. In benchmark evaluations covering general knowledge reasoning, programming, mathematics, and agent-related tasks, the K2 model outperforms other leading open-source models. Quantized at FP8.

Tool Calling
Structured Output

Added Sep 25, 2025

Model weights

Pricing

Auto routing · per 1M tokens

Input: $0.40
Output: $1.80

Compare provider prices

Specifications

Context window: 256K
Max output: 262.1K
Parameters: 1T / 32B; Total / active
Avg output (7d): 375 tokens; Longer than 34% of models

Benchmarks

Artificial Analysis

Vectara

Design Arena

Sourced from Artificial Analysis.

Intelligence Index

15.3

Better than 56% of models compared

Agentic work

T²-Bench Telecom (legacy)

Legacy fallback · Conversational AI agents in dual-control scenarios

73.4%

Better than 65% of models compared

Document reasoning

AA-LCR v1.1

Long context reasoning with updated grading

53.0%

Better than 46% of models compared

Reasoning

HLE

Humanity's Last Exam

6.4%

Better than 39% of models compared

IFBench

Instruction-following benchmark

41.7%

Better than 43% of models compared

Coding

Terminal-Bench Hard (legacy)

Legacy fallback · Agentic coding and terminal use

23.5%

Better than 63% of models compared

LiveCodeBench

Contamination-free coding benchmark

61.0%

Better than 67% of models compared

Math

AIME 2025

American Invitational Mathematics Examination 2025

57.3%

Better than 54% of models compared

Knowledge

MMLU-Pro

Professional and academic subject knowledge

81.9%

Better than 77% of models compared

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

76.7%

Better than 61% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation