Open SourceMoonshot AIReleased 2026-07

Kimi K3

Open frontier intelligence with 2.8T parameters and 1M context window

Context window

1M

Input / 1M tokens

$3.00

Output / 1M tokens

$15.00

Provider

Moonshot AI

Data verified 2026-07-18

Kimi K3 is Moonshot AI's flagship large language model, a 2.8-trillion-parameter Mixture-of-Experts model with native multimodal vision capabilities, 1M-token context window, and always-on reasoning mode. It ranks fourth overall among frontier models, trailing only Claude Fable 5 and GPT-5.6 Sol, with particularly strong coding performance.

Capability index

Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.

Coding
95
Reasoning
90
Math
85
Multimodal
85
Long context
95
Speed
70
Cost efficiency
80

How to access it

Available via Kimi API and web interface at kimi.ai. Full open-source weights scheduled for release by July 27, 2026.

Strengths

  • 1M-token context window for long-horizon tasks
  • Superior coding performance - leads Arena.ai Frontend Code benchmark
  • 2.5x improved scaling efficiency over K2
  • Multimodal (text, image, video) capabilities
  • Open-weight release planned for July 27, 2026

Best for developers who...

Coding and software engineeringLong-document analysis and researchComplex reasoning and agentic workflowsRepository-scale code navigation

When to choose it (and when not to)

Reach for Kimi K3 when...

  • When you need open-weight frontier-class model with large context
  • For coding-heavy workloads where it outperforms Fable 5
  • When planning to self-host after July 27 release

Look elsewhere if...

  • For cost-sensitive quick chat applications
  • If you need model weights immediately (not available until July 27)
  • If general knowledge tasks are priority over coding

How to use it

  • Model uses always-on maximum reasoning effort
  • Supply full context and detailed specifications for best results
  • Reason tokens are billed as output tokens

Quickstart

Python
from moonshot import MoonshotAPI
client = MoonshotAPI()
response = client.chat.completions.create(
  model='kimi-k3',
  messages=[{'role': 'user', 'content': 'Your prompt here'}]
)
print(response.choices[0].message.content)

Via Kimi API or OpenRouter/third-party providers

API model id: kimi-k3

Benchmarks

BenchmarkScoreNotes
Arena.ai Frontend Code1679 (ranked #1)Topped leaderboard within 24 hours of launch
Artificial Analysis Intelligence Index57 (Elo 1547)+732 points from K2.6, behind only Fable 5
Cost per task$0.94Similar to GPT-5.6 Sol ($1.04), half of Opus 4.8 ($1.80)

Source: Artificial Analysis, Arena.ai

Compare Kimi K3

Compare Kimi K3 with any other model

Build a comparison →
All model comparisons →

Learn the concepts