Textby Moonshot AI

Kimi K3

Kimi K3 is a 2.8T parameter open-weight multimodal reasoning model from Moonshot AI. It is suited for complex coding, knowledge work, and long-horizon agentic workflows, and is particularly strong at...

  • By price it is in the pricier half of the 347 text models GenMagic runs (312th cheapest).
  • It scores 17.4 on the Artificial Analysis Intelligence Index, 76th of the 120 text models with a score.
  • Among the models GenMagic runs, it places in the top 10 on 21 of its 48 benchmarks: 1st of 24 on Surface Evolver Bench, 3rd of 130 on Design Arena: Website, 4th of 118 on Design Arena: Data Viz, and more below.
  • It was listed on Jul 16, 2026.

Data last changed Nov 10, 2026 · tracked since Oct 29, 2026 · Every model change

Try Kimi K3

Generate with Kimi K3 right here. Verify your email to start with free credits.

Try

Specifications

Context window

1M tokens

Max output

944K tokens

Input price

$0.29 / 1M tokens

Output price

$24.00 / 1M tokens

Cached input

$0.27 / 1M tokens

Benchmarks

How Kimi K3 scores on public benchmarks. Each card links to the benchmark explainer and to the original source.

29.3

source date 2026-11-10
Source: openrouter

72

source date 2026-11-10
Source: openrouter

17.4

source date 2026-11-10
Source: openrouter
AIMEMathematics

68.9%± 7.0

lowsource date 2026-08-07
Source: epoch
APEX-AgentsAgentic & Tool Use

47.8%

unknown
Source: epoch
ARC-AGI-1Reasoning & Knowledge

94.5%

max
Source: epoch
ARC-AGI-2Reasoning & Knowledge

60.4%

max
Source: epoch
Chess PuzzlesReasoning & Knowledge

36%± 4.8

maxsource date 2026-10-13
Source: epoch
Design Arena: 3DDesign & Generation

1374 Elo

Source: designarena
Design Arena: CodeDesign & Generation

1349 Elo

Source: designarena
Design Arena: SVGDesign & Generation

1303 Elo

Source: designarena
DeepSWECoding

68.5%

max; Reasoning effort: max; Harness: mini-swe-agent
Source: epoch
DTBenchReasoning & Knowledge

91.2%

max
Source: epoch

44.2

unknown; Reasoning effort: none; Harness: mini-swe-agent
Source: epoch

39%± 7.7

maxsource date 2026-07-17
Source: epoch

72.2%± 2.7

maxsource date 2026-07-17
Source: epoch
GPQA DiamondReasoning & Knowledge

91.9%± 1.1

source date 2026-10-31
Source: openrouter
LMArena (Coding)Instruction & Preference

1505 Elo

maxsource date 2026-11-09
Source: lmarena
LMArena: Creative WritingInstruction & Preference

1452 Elo

maxsource date 2026-11-09
Source: lmarena
LMArena: Hard PromptsInstruction & Preference

1494 Elo

maxsource date 2026-11-09
Source: lmarena

1490 Elo

maxsource date 2026-11-09
Source: lmarena
LMArena: MathMathematics

1507 Elo

maxsource date 2026-11-09
Source: lmarena
LMArena: Multi-TurnInstruction & Preference

1488 Elo

maxsource date 2026-11-09
Source: lmarena
LMArena (Overall)Instruction & Preference

1474 Elo

maxsource date 2026-11-09
Source: lmarena
LMCAReasoning & Knowledge

52.7

max
Source: epoch

54.4

maxsource date 2026-09-03
Source: epoch
Mystery Game PuzzlesReasoning & Knowledge

12%± 3.3

lowsource date 2026-08-29
Source: epoch
PostTrainBenchAgentic & Tool Use

32

unknown; Scaffold: Claude Code
Source: epoch
ProofBenchMathematics

87%± 3.4

unknown
Source: epoch
SciCodeCoding

59.5%

max
Source: epoch
SimpleBenchReasoning & Knowledge

60.7%

max
Source: epoch
SimpleQA VerifiedReasoning & Knowledge

50.6%± 1.6

maxsource date 2026-08-27
Source: epoch
tau-bench (Airline)Agentic & Tool Use

73.6%± 2.8

source date 2026-10-31
Source: openrouter
WebDev ArenaAgentic & Tool Use

1674 Elo

max
Source: epoch
WeirdMLCoding

82.6%± 0.0

max
Source: epoch

Compare against every model in the full benchmark matrix.

Capabilities

Reasoning

Works through multi-step problems before answering, for harder tasks.

Vision input

Accepts images alongside text, so you can prompt with pictures and screenshots.

Tool use

Can call functions and tools, enabling agentic and structured workflows.

Structured output

Returns validated JSON that conforms to a schema you define.

Use Kimi K3 through the API

The model id is moonshotai/kimi-k3. Requests bill your GenMagic balance at the price above, and each response reports what it cost.

curl
curl https://genmagic.co/api/v1/chat/completions \
  -H "Authorization: Bearer $GENMAGIC_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moonshotai/kimi-k3",
    "messages": [{ "role": "user", "content": "Explain vector databases in two sentences." }]
  }'
Python (openai SDK)
from openai import OpenAI

client = OpenAI(base_url="https://genmagic.co/api/v1", api_key="YOUR_GENMAGIC_API_KEY")

resp = client.chat.completions.create(
    model="moonshotai/kimi-k3",
    messages=[{"role": "user", "content": "Explain vector databases in two sentences."}],
)
print(resp.choices[0].message.content)

From an AI agent, call generate_text with "model": "moonshotai/kimi-k3" on the GenMagic MCP server. Keys, limits and every endpoint are in the API documentation.

Use Kimi K3 in n8n

  1. In n8n, open the nodes panel, search for GenMagic and install it (a verified community node), then add it to your workflow.
  2. Set Resource to Text and Operation to Generate.
  3. In Model Name or ID, pick MoonshotAI: Kimi K3, or enter moonshotai/kimi-k3 as an expression.
  4. Connect your GenMagic API key (get one free). Each run is billed at the price above, and the node returns what it cost.

Ready-to-import workflows that already use the node are on GenMagic for n8n.

Similar-priced text models from other providers

Other text models GenMagic runs at a comparable price, on the same balance and API key.

ModelProviderPriceContext
MoonshotAI: Kimi K3 (this model)Moonshot AI$0.29 in / $24.00 out per 1M tokens1M
Qwen: Qwen3.8 Max PrimeQwen$6.00 in / $18.00 out per 1M tokens1M
OpenAI: GPT-5.3-CodexOpenAI$2.63 in / $21.00 out per 1M tokens400K
Amazon: Nova Premier 1.0Amazon$3.75 in / $18.75 out per 1M tokens1M
Fireworks: Ember-1Fireworks$4.50 in / $22.50 out per 1M tokens1M
Anthropic: Claude Sonnet 4.6Anthropic$4.50 in / $22.50 out per 1M tokens1M

Compare Kimi K3

More from Moonshot AI

All 9 Moonshot AI models

Price and listing history

Every change GenMagic has recorded for Kimi K3 since it began tracking the model on Oct 29, 2026. Prices are GenMagic's, the same the studio and the API charge.

DateChangeBeforeAfter
Nov 5, 2026Price down 4%$0.30 in / $24.00 out per 1M tokens$0.29 in / $24.00 out per 1M tokens
Nov 4, 2026Price up 4%$0.29 in / $22.50 out per 1M tokens$0.30 in / $24.00 out per 1M tokens
Nov 1, 2026Price up 36%$0.21 in / $16.50 out per 1M tokens$0.29 in / $22.50 out per 1M tokens
Oct 30, 2026Updated: coding index, agentic index, intelligence index--

In our guides

The guides that discuss Kimi K3 or show it in a live table.

Where these figures come from

  • Price: GenMagic's own price, what the studio and the API charge, recomputed from the provider's price list every few minutes and recorded daily. How pricing works.
  • Specs and description: as Moonshot AI publishes them (model card on Hugging Face), checked daily.
  • Benchmarks: 48 scores from Artificial Analysis and OpenRouter, Epoch AI, Design Arena, LMArena, re-read daily. Each score above links to its source.
  • Changes: the data on this page last changed on Nov 10, 2026, tracked since Oct 29, 2026. Every model's new listings, price changes and retirements are in the model changes log, and how every figure on GenMagic is sourced is on our sources page.

Generate with Kimi K3 today

One GenMagic plan unlocks Kimi K3 and every other model across text, image, audio, and video.