NOUMENOpen app
NOUMEN

Moonshot · Open weights

Kimi K3

Added Jul 16, 2026Open weights

Compare models

Choose up to 3 models to compare.1 of 3 chosen

Loading models…

Use in the app

Kimi K3 is Moonshot AI's open-weight, always-thinking multimodal model for long-context reasoning, coding, tool use, and native image/video understanding.

  • Reasoning
  • Tools
  • Vision
  • Structured output
  • Video input
Input
Text, Images, Video
Output
Text
Reasoning effort
low · high · max
Discovery tier
Open weights

Pricing

USD per 1M tokens
Input
$1.20
Output
$5.50
Cache read
$0.20

Reference prices as published by the model provider. API example

Specifications

Context window
1M
Max output
—
Average output (7 days)
1.1K tokens
Longer than 75% of 473 models

Side by side

Kimi K3 compared

Specifications, prices, speed and benchmark scores next to each other. Benchmarks below show every compared model too.

Compare models

Choose up to 3 models to compare.1 of 3 chosen

Loading models…

Kimi K3MoonshotThis modelStep 5 PreviewstepfunRemove
ProviderMoonshotstepfun
Discovery tierOpen weightsBalanced
AddedJul 16, 2026Sep 20, 2026
Limits
Context window1.05MBest1M
Average output1.1K tokens1.5K tokens
Pricing, USD per 1M tokens
Input$1.20$1.00Best
Output$5.50$2.70Best
Cache read$0.20$0.05Best
Performance
Throughput194.8 tok/s—
Time to first token4.0 s—
Request success (24h)99.9%100.0%Best
Benchmark indexes
Intelligence Index43.643.7Best
Coding Index76.2—
Capabilities
ReasoningYesYes
ToolsYesYes
VisionYesYes
Structured outputYesYes
Video inputYesYes
Parallel toolsNoYes
Reasoning effortlow, high, maxlow, medium, high

Public evaluations

Benchmarks

Independent evaluation results, with how each score ranks among the models in the Noumen catalog.

  • Kimi K3
  • Step 5 Preview

Intelligence Index

  • Kimi K343.6
  • Step 5 Preview43.7

Kimi K3 tested at max reasoning effort

Kimi K3: Better than 89% of 326 scored models

3.8Catalog models by score55.1

Indexes

  • Coding Index

    Composite of the coding evaluations

    • Kimi K3: 76.2
    • Step 5 Preview: —

Agentic work

  • AutomationBench-AA

    Workflow automation, penalized for guardrail violations

    • Kimi K3: 58.3%Top
    • Step 5 Preview: 51.0%

  • AutomationBench-AA tasks completed

    Workflows finished end to end without a guardrail violation

    • Kimi K3: 24.8%Top
    • Step 5 Preview: 17.7%

  • Harvey LAB-AAv1.1

    Legal tasks passed in full, with no material hallucination

    • Kimi K3: 5.3%
    • Step 5 Preview: —

  • AA-Briefcasev1.1

    Agentic knowledge work, rated head to head

    • Kimi K3: 1,499 EloTop
    • Step 5 Preview: 1,423 Elo

  • GDPval-AAv2.1

    Economically valuable occupational tasks, rated head to head

    • Kimi K3: 1,534 Elo
    • Step 5 Preview: 1,587 EloTop

Document reasoning

  • GDP.pdf

    Professional reasoning over PDF documents, all-pass rate

    • Kimi K3: 22.0%Top
    • Step 5 Preview: 14.8%

  • AA-LCRv1.1

    Reasoning across long documents

    • Kimi K3: 88.7%Top
    • Step 5 Preview: 88.3%

  • MLCR-AA

    Long-context reasoning over medical records

    • Kimi K3: 21.7%Top
    • Step 5 Preview: 18.3%

Reasoning and knowledge

  • Humanity’s Last Exam

    Expert-level questions across academic fields

    • Kimi K3: 46.9%Top
    • Step 5 Preview: 46.5%

  • GPQA Diamond

    Graduate-level science questions

    • Kimi K3: 93.5%
    • Step 5 Preview: —

Coding

  • Terminal-Benchv4.0

    Practical software engineering in a terminal

    • Kimi K3: 12.6%
    • Step 5 Preview: 33.3%Top

  • SciCode

    Scientific computing in Python

    • Kimi K3: 59.5%Top
    • Step 5 Preview: 58.9%

Legacy evaluations

  • AA-LCR (unversioned)

    Long-context reasoning, earlier grading

    • Kimi K3: 88.7%Top
    • Step 5 Preview: 88.3%

Scores from Artificial Analysis, updated Oct 26, 2026. Kimi K3 was tested at max reasoning effort; compare scores measured at the same effort.Artificial Analysis

Live routing data

Performance and reliability

Recent averages across requests routed for this model. Your own speed depends on prompt length, reasoning effort and load.

Throughput
194.8 tok/s
Faster than 84% of 256 models
Time to first token
4.0 s
Quicker than 38% of 256 models
Prompt cache hits
84.7%
About $0.35 per 1M input tokens with caching
Request success (24h)
99.9%
Reliable

Cache hits are the share of input tokens recently served from a prompt cache across all requests. They are not a forecast of your own hit rate.

Hourly request success · last 72 hoursLowest hour 98.6%
  • Healthy (98% or more)
  • Degraded (90–98%)
  • Down (under 90%)
  • No requests

Developers

Use it through the API

Noumen’s API is OpenAI-compatible: any Chat Completions client works with a base URL and a key.

Read the quickstart
curl https://www.noumen.xyz/api/v1/chat/completions \
  -H "Authorization: Bearer $NOUMEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moonshotai/kimi-k3",
    "reasoning_effort": "low",
    "messages": [{ "role": "user", "content": "Explain this stack trace." }]
  }'
import os
from openai import OpenAI

client = OpenAI(base_url="https://www.noumen.xyz/api/v1", api_key=os.environ["NOUMEN_API_KEY"])
reply = client.chat.completions.create(
    model="moonshotai/kimi-k3",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

Moonshot

Other Moonshot models, starting with the same family.