MoonshotAI: Kimi K3

Moonshot

open_source_flagship

Capabilities

VisionToolsStructured OutputExtended ThinkingLong ContextCode

Strengths

Exceptional code generation and reasoning (0.96 fitness)
Massive 1M token context for long-document workflows
Open weights enable full data privacy via self-hosting
Strong vision-language integration for multimodal tasks
Extended thinking and structured output support

Weaknesses

Very slow inference (0.03 speed score, ~33 tok/s)
High output pricing ($15/M) limits cost-effective scaling
Large model size requires substantial compute for inference
Training data composition not fully disclosed
Sustainability footprint of large model not quantified

Pricing

Input / 1M tokens

Output / 1M tokens

$15

Context window

1,048.576k

Transparency

Open weights

10.0 / 10

Open training data

0.0 / 10

Open methodology

7.5 / 10

Licence openness

9.0 / 10

Provider disclosure

6.5 / 10

FMTI company score

N/A

Composite:6.0 / 10

Open-weight model with strong public documentation and Apache 2.0 license. Moonshot discloses model architecture and capabilities but training data composition and methodology less detailed than Meta. Weights publicly available for self-hosting.

Sustainability

Inference energy

N/A

Training footprint

N/A

Provider infrastructure

5.0 / 10

Composite:5.0 / 10

Large 2.8T parameter model implies significant inference energy cost. No public commitments on renewable energy or carbon offsetting from Moonshot. Provider infrastructure sustainability unknown.

Task Fitness

question answering

91%

code

96%

mathematics

50%

business communication

50%

vision

78%

analysis

92%

extraction

90%

long-form generation

86%

research

50%

reasoning

92%

summarisation

92%

translation

85%

conversation

88%