Skip to content
AITrendTool

Kimi

Moonshot AI's assistant on open-weight K2 models, with agentic coding and a 256K context window

Kimi is Moonshot AI's assistant built on its open-weight K2 models — headlined by K2.6 and the K2.7 Code agent, both with a 262,144-token (256K) context window. The free Adagio tier covers light use; paid plans run $19 to $199/month, and the developer API runs $0.95 per million input tokens and $4 per million output. Best for agentic coding, deep research, and long-document work on a budget.

Verified JUN 26, 2026 FREEMIUM Live
Visit kimi.com ↗ AI Chatbots

What is Kimi?

Kimi is an AI assistant from Moonshot AI, a Chinese lab, built around its open-weight K2 model family — a roughly one-trillion-parameter Mixture-of-Experts design. The headline models are Kimi K2.6 for general tasks and Kimi K2.7 Code for agentic software engineering, both offering a 262,144-token context window that can hold an entire codebase or a book-length document at once. Beyond plain chat, Kimi bundles agentic tooling: Deep Research for multi-step web synthesis, Kimi Code for long-horizon coding, an Agent Swarm that runs parallel subagents, and productivity agents that generate slides, sheets, and simple websites.

Because the K2 weights are open, Kimi is both a product and a model provider. You can use the hosted assistant through consumer membership tiers, or call the models directly through a token-based developer API whose cache-hit input pricing is unusually low. That dual nature makes Kimi attractive to cost-sensitive developers who want frontier-class agentic coding and long-context reasoning without paying the rates of the largest Western providers. The trade-off is that pricing and availability vary by region and channel, and the consumer plans meter usage through quotas rather than offering flat, unlimited access.

Who is it for?

Kimi fits users who value long context, agentic features, and low cost, and who are comfortable with a China-based provider. The free Adagio tier is enough to evaluate the assistant, but serious agentic or coding work pushes toward the paid memberships or the API.

  • Cost-sensitive developers who want agentic coding and a 256K context window without top-tier Western pricing.
  • Researchers and analysts running Deep Research over long documents or cross-language sources.
  • Builders and tinkerers who want to call open-weight K2 models through an inexpensive token API.
  • Power users who can work within quota-metered plans and accept region-dependent pricing and compliance considerations.

How much does Kimi cost?

Starting price: $19/mo · Free tier: yes · Model: freemium

Pricing verified JUN 26, 2026

Price history tracked from June 2026

Kimi pricing tiers, verified against the official pricing page
Plan Price Includes
Adagio Free 6 agent task quota · 1 concurrent task · 200 professional database calls
Moderato $19/mo 60 agent task quota, 2 concurrent tasks · 25 Agent Swarm uses · Kimi Code access, 4x speed priority · 2,000 database calls · $15/mo billed annually
Allegretto $39/mo 150 agent task quota · 50 Agent Swarm uses, 5x Kimi Code · Kimi Claw access · 5,000 database calls
Allegro $99/mo 360 agent task quota, 4 concurrent tasks · 120 Agent Swarm uses, 15x Kimi Code · 12,000 database calls
Vivace $199/mo 720 agent task quota · 240 Agent Swarm uses, 30x Kimi Code · 24,000 database calls

What are Kimi's key features?

  • Open-weight Kimi K2 model family — a 1-trillion-parameter Mixture-of-Experts architecture
  • Kimi K2.6 general model and K2.7 Code agentic coding model
  • 262,144-token (256K) context window on both flagship models
  • Kimi Code for long-horizon, agentic software engineering
  • Deep Research and web search for multi-step source synthesis
  • Agent Swarm — parallel subagents that scale with plan tier
  • Productivity agents for slides, sheets, and websites
  • Token-based developer API with cheap cache-hit input pricing

What people use Kimi for

  1. 01 Running an agentic coding task across a large codebase with Kimi Code and a 256K-token context
  2. 02 Summarizing and cross-referencing long documents, reports, or research papers in one session
  3. 03 Multi-step web research with Deep Research that gathers and synthesizes sources
  4. 04 Generating slides, sheets, and simple websites from a prompt with the built-in productivity agents
  5. 05 Calling the K2 models via API for cost-sensitive bulk generation or classification

Pros and cons

Pros and cons of Kimi
Pros Cons
Open-weight K2 models can be self-hosted or accessed through third parties, reducing lock-in Consumer pricing varies by region and channel — Moonshot itself warns figures differ across web, app, and country — making true cost hard to pin down
256K-token context handles whole codebases and long documents in one pass Paid plans meter agent quota, concurrent tasks, Agent Swarm uses, and database calls, so heavy users hit caps even after paying
API pricing is low — $0.95 per million input tokens, with cache hits as cheap as $0.16 As a China-based lab's product, data residency, compliance, and content moderation may concern some Western or enterprise buyers
Free Adagio tier lets you try agentic features before paying

What are the best Kimi alternatives?

See all Kimi alternatives →

How people make money with Kimi

  • Deliver fixed-scope feature builds for small teams using Kimi Code and its 256K context, sold as milestone gigs on Upwork or Contra
  • Offer long-document analysis and Chinese-English market research with Kimi's Deep Research mode as a freelance analyst service

Frequently asked questions

Is Kimi free?

Kimi has a free tier called Adagio that includes a 6-task agent quota, one concurrent task, and 200 professional database calls. It is enough to try the assistant and agentic features, but heavier use requires a paid membership starting at $19 per month.

What models does Kimi use?

Kimi runs on Moonshot AI's open-weight K2 family, a trillion-parameter Mixture-of-Experts architecture. The current flagships are Kimi K2.6 for general use and Kimi K2.7 Code for agentic coding, both offering a 262,144-token context window.

How much does Kimi cost?

Consumer plans run from $19 per month for Moderato up to $199 per month for Vivace, with discounts for annual billing. Separately, the developer API charges per token: about $0.95 per million input tokens and $4 per million output tokens.

Does Kimi have an API?

Yes. Moonshot AI offers a token-based developer API for the K2 models at its platform. Pricing is per million tokens, with heavily discounted cache-hit input rates that make repeated-context workflows cheaper than standard input pricing.

What is Kimi's context window?

Both Kimi K2.6 and K2.7 Code support a 262,144-token context window, roughly 256K tokens. That is large enough to load an entire codebase or a book-length document into a single session without losing earlier context.

Is Kimi good for coding?

Kimi K2.7 Code is built for agentic, long-horizon software engineering, and the large context window helps it reason over big codebases. For pure coding quality, many developers still rank Claude higher, but Kimi is competitive at a lower price.

Public signals