AI Models Navi
Back to Models
Moonshot AIProprietary

Kimi K2 Thinking

Kimi K2 Thinking is an inference model developed by Moonshot AI. It features a large model configuration with approximately 10.4 trillion parameters and a 256K long context window.

Parameters

10400.0B

Context Window

256K

License

https://github.com/openai/gpt-2/blob/master/LICENSE

Release Date

2025-11-06

API Pricing

API pricing for this model is not yet available

Strengths

  • ・Massive over 10 trillion parameter count
  • ・Wide 256K context window
  • ・Specialized for advanced reasoning capabilities

Weaknesses

  • ・Extremely large model size over 1TB
  • ・Closed distribution format
  • ・Increased resource consumption

Use Cases

  • ・Executing complex logical reasoning
  • ・Analyzing and processing large document sets
  • ・Advanced problem-solving tasks

Deep Analysis

Architecture

MoE (1T total, 32B active)

Mixture-of-Experts reasoning model

Context Window

256K tokens

API Pricing (Input/Output)

$0.15 / $0.60 per 1M tokens

Via Kimi API Platform

Release Date

November 2025

Training Cost

$4.6M

Reported by Implicator

Tool Calls

Up to 300 sequential

Agentic capability

Strengths

  • ・Outperforms GPT-5 on HLE reasoning benchmark
  • ・Open-weight model
  • ・Supports up to 300 sequential tool calls for agentic tasks
  • ・Competitive pricing at $0.15/$0.60 per 1M tokens
  • ・Strong deep reasoning capabilities

Weaknesses

  • ・No vision support
  • ・Discontinued May 25 2026 (superseded by kimi-k2.6)
  • ・Higher latency than turbo variants
  • ・Limited to text-only modality

Competitor Comparison

ModelArenaSWEGPQAPrice
GPT-5---Higher
Claude Opus 4.5---Higher
DeepSeek V3.2---Comparable

Sources

Analysis generated: 2026-05-24