NOUMENOpen app
NOUMEN

Moonshot · Open weights

Kimi K2.5 Thinking

Added Jan 26, 2026Open weights

Compare models

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

Loading models…

Use in the app

Kimi K2.5 with thinking mode enabled. Built on Kimi K2 with ~15T mixed visual and text tokens, it excels at general reasoning, visual coding, and agentic tool-calling. Produces reasoning traces for complex multi-step workflows.

  • Reasoning
  • Tools
  • Vision
  • Parallel tools
Input
Text, Images
Output
Text
Discovery tier
Open weights

Pricing

USD per 1M tokens
Input
$0.38
Output
$2.00
Cache read
$0.07

Reference prices as published by the model provider. API example

Specifications

Context window
262.1K
Max output
65.5K
Average output (7 days)
2.4K tokens
Longer than 93% of 465 models

Side by side

Kimi K2.5 Thinking 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 K2.5 ThinkingMoonshotThis modelKimi K2.6MoonshotRemove
ProviderMoonshotMoonshot
Discovery tierOpen weightsOpen weights
AddedJan 26, 2026Apr 16, 2026
Limits
Context window262.14K262.14K
Max output65.54K65.54K
Average output2.4K tokens594 tokens
Pricing, USD per 1M tokens
Input$0.38Best$0.45
Output$2.00$1.89Best
Cache read$0.07Best$0.0757
Performance
Throughput30.3 tok/s46.6 tok/sBest
Time to first token2.7 sBest2.7 s
Request success (24h)99.5%99.7%Best
Benchmark indexes
Intelligence Index23.527.0Best
Coding Index46.861.8Best
Capabilities
ReasoningYesNo
ToolsYesYes
VisionYesYes
Parallel toolsYesNo
Reasoning effort——

Public evaluations

Benchmarks

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

  • Kimi K2.5 Thinking
  • Kimi K2.6

Intelligence Index

  • Kimi K2.5 Thinking23.5
  • Kimi K2.627.0

Kimi K2.5 Thinking: Better than 57% of 326 scored models

3.8Catalog models by score55.1

Indexes

  • Coding Index

    Composite of the coding evaluations

    • Kimi K2.5 Thinking: 46.8
    • Kimi K2.6: 61.8Top

Agentic work

  • AutomationBench-AA

    Workflow automation, penalized for guardrail violations

    • Kimi K2.5 Thinking: —
    • Kimi K2.6: 13.0%

  • AutomationBench-AA tasks completed

    Workflows finished end to end without a guardrail violation

    • Kimi K2.5 Thinking: —
    • Kimi K2.6: 1.8%

  • AA-Briefcasev1.1

    Agentic knowledge work, rated head to head

    • Kimi K2.5 Thinking: —
    • Kimi K2.6: 819 Elo

  • GDPval-AAv2.1

    Economically valuable occupational tasks, rated head to head

    • Kimi K2.5 Thinking: 839 Elo
    • Kimi K2.6: 1,038 EloTop

Document reasoning

  • GDP.pdf

    Professional reasoning over PDF documents, all-pass rate

    • Kimi K2.5 Thinking: —
    • Kimi K2.6: 13.0%

  • AA-LCRv1.1

    Reasoning across long documents

    • Kimi K2.5 Thinking: 78.0%
    • Kimi K2.6: 81.0%Top

  • MLCR-AA

    Long-context reasoning over medical records

    • Kimi K2.5 Thinking: —
    • Kimi K2.6: 11.1%

Reasoning and knowledge

  • Humanity’s Last Exam

    Expert-level questions across academic fields

    • Kimi K2.5 Thinking: 30.7%
    • Kimi K2.6: 37.5%Top

  • GPQA Diamond

    Graduate-level science questions

    • Kimi K2.5 Thinking: 87.9%
    • Kimi K2.6: 91.1%Top

  • IFBench

    Following precise output instructions

    • Kimi K2.5 Thinking: 70.2%
    • Kimi K2.6: 76.0%Top

Coding

  • Terminal-Benchv4.0

    Practical software engineering in a terminal

    • Kimi K2.5 Thinking: —
    • Kimi K2.6: 0.5%

  • SciCode

    Scientific computing in Python

    • Kimi K2.5 Thinking: —
    • Kimi K2.6: 51.5%

Legacy evaluations

  • Terminal-Bench Hard

    Earlier terminal task suite

    • Kimi K2.5 Thinking: 34.8%
    • Kimi K2.6: 43.9%Top

  • τ²-Bench Telecom

    Tool use in a customer-support simulation

    • Kimi K2.5 Thinking: 95.9%Top
    • Kimi K2.6: 95.9%Top

  • AA-LCR (unversioned)

    Long-context reasoning, earlier grading

    • Kimi K2.5 Thinking: 78.0%
    • Kimi K2.6: 81.0%Top

Scores from Artificial Analysis, updated Oct 26, 2026. 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
30.3 tok/s
Faster than 20% of 250 models
Time to first token
2.7 s
Quicker than 63% of 250 models
Prompt cache hits
59.4%
About $0.19 per 1M input tokens with caching
Request success (24h)
99.5%
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 97.2%
  • 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-k2.5:thinking",
    "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-k2.5:thinking",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

Moonshot

Other Moonshot models, starting with the same family.