Kimi K2-Thinking-0905

Name: Kimi K2-Thinking-0905
Author: Moonshot AI

Moonshot AI

Kimi K2 Thinking 0905 is the September 2025 update to the reasoning-focused variant of Kimi K2. It is a large-scale Mixture-of-Experts (MoE) language model by Moonshot AI with 1 trillion parameters and 32 billion active per forward pass. The thinking variant is specifically optimized for complex reasoning, mathematical proofs, and multi-step problem solving. This update enhances agentic coding, frontend development, and supports long-context inference up to 256K tokens.

Key Specifications

Parameters

1.0T

Context

262.1K

Release Date

September 4, 2025

Average Score

68.3%

API Documentation Research Paper Repository Model Weights Results Blog

Timeline

Key dates in the model's history

Announcement / Last Update

September 4, 2025

Today

September 11, 2026

Technical Specifications

Parameters

1.0T

Training Tokens

15.5T tokens

Knowledge Cutoff

Family

Capabilities

MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)

$0.60

Output (per 1M tokens)

$2.40

Max Input Tokens

262.1K

Max Output Tokens

65.5K

Supported Features

Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

Programming

Programming skills tests

SWE-Bench Verified

w/ tools • Self-reported

71.3%

Reasoning

Logical reasoning and analysis

GPQA

Diamond • Self-reported

84.5%

Other Tests

Specialized benchmarks

AIME 2025

Standard evaluation • Self-reported

100.0%

HMMT 2025

Standard evaluation • Self-reported

97.5%

MMLU-Redux

Without tools • Self-reported

94.0%

MMLU-Pro

Without tools • Self-reported

85.0%

BrowseComp

w/ tools • Self-reported

60.2%

BrowseComp-zh

w/ tools • Self-reported

62.3%

FinSearchComp-T3

w/ tools • Self-reported

47.4%

FRAMES

w/ tools • Self-reported

87.0%

HealthBench

no tools • Self-reported

58.0%

Humanity's Last Exam

heavy • Self-reported

51.0%

IMO-AnswerBench

no tools • Self-reported

78.6%

LiveCodeBench v6

no tools • Self-reported

83.1%

Multi-SWE-Bench

w/ tools • Self-reported

41.9%

OJBench

cpp, no tools • Self-reported

48.7%

SciCode

no tools • Self-reported

44.8%

Seal-0

w/ tools • Self-reported

56.3%

SWE-bench Multilingual

w/ tools • Self-reported

61.1%

Terminal-Bench

w/ simulated tools (JSON) • Self-reported

47.1%

WritingBench

Longform Writing, no tools • Self-reported

73.8%

License & Metadata

License

mit

Announcement Date

September 4, 2025

Last Updated

September 4, 2025

Similar Models

All Models

Kimi K2 Instruct

Moonshot AI

1.0T

Best score:0.9 (HumanEval)

Released:Jan 2025

Price:$0.57/1M tokens

Kimi K2-Instruct-0905

Moonshot AI

1.0T

Best score:0.9 (MMLU)

Released:Sep 2025

Price:$0.60/1M tokens

Kimi K2 0905

Moonshot AI

1.0T

Best score:0.9 (HumanEval)

Released:Sep 2025

Price:$0.60/1M tokens

Kimi K2 Base

Moonshot AI

1.0T

Best score:0.9 (MMLU)

Released:Jan 2025

Kimi K2.6

Moonshot AI

MM1.0T

Best score:0.9 (GPQA)

Released:Apr 2026

Price:$1.20/1M tokens

Kimi K2.5

Moonshot AI

MM1.0T

Best score:0.9 (GPQA)

Released:Jan 2026

GLM-4.5-Air

Zhipu AI

106.0B

Best score:0.8 (TAU)

Released:Jul 2025

Hy3

Tencent

295.0B

Best score:0.9 (GPQA)

Released:Jul 2026

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.