AI Models Navi
Back to Models
Moonshot AIOpen Source

Kimi K2-Instruct-0905

A foundation model developed by Moonshot AI, boasting a large number of parameters. It supports a long context window of 256K and is designed as a chat-specialized model with advanced dialogue capabilities.

Parameters

10000.0B

Context Window

256K

License

MIT

Release Date

2025-09-05

API Pricing

API pricing for this model is not yet available

Strengths

  • ・Massive parameter count at 10 trillion scale
  • ・Expansive 256K context window
  • ・High freedom with MIT license

Weaknesses

  • ・Enormous model size over 1TB
  • ・Requires very high computational resources
  • ・Potentially high operational costs

Use Cases

  • ・Analysis of ultra-long documents
  • ・Dialogue systems requiring complex context
  • ・Knowledge extraction from large-scale data

Deep Analysis

Architecture

MoE (1T total, 32B active)

Enhanced agentic coding variant

Context Window

256K tokens

Training Data

15.5T tokens

Release Date

September 2025

License

Open-weight

Available on HuggingFace

Strengths

  • ・Enhanced agentic coding abilities over base K2
  • ・Improved frontend code quality
  • ・Better context understanding
  • ・Open-weight on HuggingFace
  • ・Strong tool use and function calling

Weaknesses

  • ・No vision support
  • ・Large model size
  • ・Superseded by newer K2.5 and kimi-k2.6

Competitor Comparison

ModelArenaSWEGPQAPrice
GPT-4o---Higher
Claude Sonnet 4---Higher
DeepSeek V3---Comparable

Kimi K2-Instruct-0905 is an enhanced version of Kimi K2 focused on agentic coding, with improved frontend code quality and better context understanding. It builds on the 0711-preview with targeted improvements for software engineering workflows.

Sources

Analysis generated: 2026-05-24