Models Model details

MoonshotAI

MoonshotAI: Kimi K2.5

moonshotai/kimi-k2.5

Kimi K2.5 is Moonshot AI's native multimodal model, delivering state-of-the-art visual coding capability and a self-directed agent swarm paradigm. Built on Kimi K2 with continued pretraining over approximately 15T mixed visual and text tokens, it delivers strong performance in general reasoning, visual coding, and agentic tool-calling.

Input
$0.5

/ 1M tokens

Output
$2.5

/ 1M tokens

Context window
262.14K

tokens

Limits

Context

262,144

tokens

Max output

235,929

90% of the context window

Parameters

1,026,879,376,368

Model facts

Author

MoonshotAI

Tokenizer

Other

Modality

Text + Image -> Text

License

Other

Created

Jan 27, 2026

Hugging Face

moonshotai/Kimi-K2.5

Publication status

Hugging Face: verifiedLicense: publishedParameter count: published

Modalities

Input · TextInput · ImageOutput · Text

Generation controls

Frequency penaltyInclude reasoningLogit biasLogprobsMax tokensMin pPresence penaltyReasoningRepetition penaltyResponse formatSeedStopStructured outputsTemperatureTool choiceToolsTop kTop logprobsTop p

Capability scores

46.8
Coding index

Better than 43% of models compared

23.5
Intelligence index

Better than 57% of models compared

Agent tasks

Category performance

Godot games
Top 48%

1219 Elo

59.8% win

Android apps
Top 71%

1133 Elo

57.8% win

Mobile apps
Top 66%

1145 Elo

54.7% win

Full-stack apps
Top 66%

1114 Elo

54.5% win

Web apps
Top 68%

1129 Elo

51.6% win

Creative tasks

Category performance

Websites
Top 37%

1250 Elo

55.2% win

Code
Top 41%

1238 Elo

54.1% win

UI components
Top 44%

1236 Elo

53.6% win

Games
Top 44%

1208 Elo

53.4% win

3D
Top 44%

1206 Elo

53.1% win

Data visualization
Top 47%

1220 Elo

51.3% win

SVG
Top 59%

1149 Elo

48.4% win

ASCII art
Top 54%

1168 Elo

46.4% win