Kimi K3

Our take on Kimi K3

Kimi K3: Moonshot's 2.8T open model, top open model on LMArena. Best for best-in-class open weights for chat and coding. Watch out for: huge (2.8t parameters, server only); custom "kimi k3" license; official api price not verified. Price: Official API price not verified (about $2.30 per 1M blended per Artificial Analysis) (no free tier, checked Oct 2026).

What Kimi K3 is good at

• Chat: LMArena Text. #18 of ~410 on LMArena Text (kimi-k3-max), 2026-10-15
• Coding: LMArena Coding. #12 of ~410 on LMArena Coding (kimi-k3-max), 2026-10-15
• Writing: LMArena Creative Writing. #27 of ~410 on LMArena Creative Writing (kimi-k3-max), 2026-10-15
• Agents & tool use: Artificial Analysis Intelligence Index. 44 on the AA Intelligence Index v4.3.2 (Kimi K3 (max)), a composite that includes agentic and tool-use evals, 2026-10-15

Whole scores from 1 to 5, mapped from the source shown. “Not rated” means we found no public benchmark for that job; “not supported” means the model can’t do it. How the bars are scored.

What Kimi K3 costs

From artificialanalysis.ai/models/open-source, checked 2026-10-15, unverified. Some prices could not be read from the official page, so confirm before buying. Prices are before tax and vendors change them often.

What outside sources say

Our scores: quality 3.93/5, ease 3/5, value 2.5/5. Editorial start 4/3; benchmark adjustment -0.07, rating adjustment +0. How we rank.

If Kimi K3 isn’t the one

• 01

  Ollama

  From $20/mo, free tier first.
  Run open AI models on your own computer, free. Seeks private ChatGPT-style chat with open models, offline.
  Add
• 02

  Nous Portal

  From $20/mo, free tier first.
  One Nous sign-in for 200+ models, hosted tools and agent hosting. Seeks hermes Agent users who want one bill instead of a key per lab.
  Add
• 03

  OpenRouter

  Paid price unverified, free tier first.
  One API key for 500+ models at provider list prices. Seeks trying and switching between many models without separate accounts.
  Add
• 04

  Gemini 4 Argon

  Paid price unverified.
  Google's flagship reasoning model, top of LMArena right now. Seeks hard reasoning, coding and long agent runs through the Gemini API.
  Add
• 05

  Claude Opus 5.5

  Paid price unverified.
  Anthropic's daily-driver flagship, #1 on Artificial Analysis. Seeks agentic coding, careful writing and long multi-step work.
  Add
• 06

  GPT-6.1 Sol

  Paid price unverified.
  OpenAI's mid-priced flagship with a 1M-token window. Seeks strong all-rounder for coding and agents at a lower price than GPT-6 Astra.
  Add

Questions people ask