Description
Kimi is Moonshot AI’s all-in-one agentic AI workspace, combining chat, Deep Research, document and spreadsheet tools, a website builder, and browser and desktop agents under one product at kimi.com. Its current flagship model, Kimi K3, is a 2.8 trillion-parameter, open-weight model with a 1 million-token context window and native vision support, positioned by Moonshot as frontier-level though trailing the very top proprietary models. Five membership tiers, Adagio through Vivace, scale usage: agent concurrency, scheduled tasks, Agent Swarm subagents, and Kimi Code credits all increase with tier, while the full 1M-token extra-long chat capacity only unlocks on Allegro and Vivace. Demand following K3’s July 2026 launch pushed Moonshot’s GPU capacity to its limit within 48 hours, leading the company to temporarily pause new paid subscriptions while reopening spots in batches; existing subscribers were unaffected. Moonshot has also announced it will split membership into a separate Kimi (web/app/work) plan and a dedicated Kimi Code plan. It suits budget-conscious users who want a capable, long-context AI workspace at a lower price than comparable Western subscriptions.
Key Features
- Kimi K3 model — 2.8 trillion-parameter open-weight flagship with a 1 million-token context window and native vision.
- Agent Swarm — runs multiple subagents in parallel on the same task, scaling from 2 subagents (Moderato) to 8 (Allegro/Vivace).
- Kimi Work — local desktop AI agent for knowledge work, separate from the browser-based chat.
- Kimi Claw — one-click cloud deployment for running AI agents continuously.
- Deep Research, Slides, Sheets, and Docs — built-in tools for research reports, presentations, spreadsheets, and documents.
- Kimi Code credits — coding agent usage scales from 1x (Moderato) up to 30x (Vivace) credits.
- Automatic context caching — repeated context is billed at a 90% discount on the API, reducing cost for long-running sessions.
How It Works
Kimi runs as a web and mobile chat workspace where users interact with the Kimi K3 model for everyday questions, research, and agentic tasks. Higher membership tiers unlock more concurrent agent tasks, faster priority queueing, and access to Agent Swarm, which splits a task across multiple subagents working in parallel rather than one model working sequentially. Kimi Work extends the same capability to a local desktop agent, while Kimi Claw allows one-click deployment of an agent that runs continuously in the cloud rather than only during an active session. The full 1M-token extra-long chat capacity, useful for entire codebases or lengthy documents, is reserved for the Allegro and Vivace tiers; lower tiers use K3 with a smaller effective context. Kimi Code, Moonshot’s coding-specific product, draws from the same membership credit pool, with usage multipliers scaling from 1x on the entry paid tier up to 30x on the top tier.
Technical Architecture & Overview
- Core Engine: Kimi K3, a 2.8 trillion-parameter open-weight model with Stable LatentMoE architecture (16 of 896 experts active per token) and a 1,048,576-token context window, per Moonshot’s own technical documentation.
- Deployment: Web and mobile chat app, Kimi Work desktop agent, Kimi Claw cloud deployment, and Kimi Code CLI for terminal and IDE use; full model weights were released publicly on July 26, 2026.
- API Surface: Kimi K3 API at platform.kimi.ai, billed per million tokens ($3.00 input on cache miss, $0.30 on cache hit, $15.00 output), separate from and not included in membership plans.
- Known Limits: New paid subscriptions were temporarily paused starting July 19, 2026 after demand pushed GPU capacity to its limit within 48 hours of K3’s launch; Moonshot is reopening spots in batches, and existing subscribers were not affected.
Pros & Cons
| Pros | Cons |
|---|---|
| Kimi K3’s API pricing undercuts comparable Western frontier models by a wide margin on a per-token basis, per Moonshot’s own published rate card | New paid subscriptions were temporarily paused as of July 19, 2026 due to GPU capacity constraints following K3’s launch |
| Full model weights are open and publicly downloadable, unlike most comparably capable frontier models | The full 1M-token extra-long chat capacity is reserved for the Allegro ($79/month) and Vivace ($159/month) tiers only |
| Automatic context caching cuts input token costs by 90% on repeated context, reducing cost for long-running sessions | Moonshot has announced it will split membership into separate Kimi and Kimi Code products, so current plan details may change |
| Agent Swarm and Kimi Claw extend the platform into genuine parallel and always-on agent execution, not just single-session chat | Kimi Code, chat, and other agent usage all draw from one shared membership credit pool, so heavy use of one can reduce capacity for the others |
| Free Adagio tier includes real agent access, not just basic chat, though with tight limits | Kimi K3’s overall performance is described by Moonshot itself as trailing the top proprietary models, even while leading most other tested models |
Pricing
| Plan | Price |
|---|---|
| Adagio | $0/month |
| Moderato | $15/month billed annually ($19/month billed monthly) |
| Allegretto | $31/month billed annually ($39/month billed monthly) |
Platform Availability
Web | iOS | Android | Desktop (Kimi Work) | API
Best For
Budget-conscious users | Long-context research and coding | Agentic automation
Frequently Asked Questions
Is Kimi AI free to use?
Yes. The Adagio tier is free and includes real agent access, though with tight limits on concurrent tasks and no Agent Swarm or Kimi Code, according to Moonshot’s official membership pricing page.
Can I sign up for a paid Kimi plan right now?
It depends. Moonshot temporarily paused new paid subscriptions starting July 19, 2026 after demand for Kimi K3 pushed its GPU capacity to the limit within 48 hours of launch; the company is reopening spots in batches, per its own official statement.
Which plans include the full 1M-token context window?
Only the Allegro and Vivace tiers unlock K3’s extra-long chat capacity of up to 1M tokens; lower tiers use the same K3 model at a smaller effective context, according to Moonshot’s official pricing page.
Is Kimi K3 open source?
Yes. Moonshot released Kimi K3’s full model weights publicly on July 26, 2026, making it, according to widely reported benchmarks, one of the largest open-weight models publicly available at launch.
How does Kimi AI differ from Kimi Code?
Kimi AI is the general chat, research, and agent workspace at kimi.com, while Kimi Code is Moonshot’s dedicated terminal and IDE coding agent; both currently draw from the same membership credit pool, though Moonshot has announced plans to split them into separate subscriptions.
Similar Tools in This Directory
Also in this directory: ChatGPT, OpenAI’s assistant with bundled Codex access | Claude by Anthropic, coding agent bundled into every paid plan | Google Gemini AI, multimodal assistant integrated into Google Workspace | Meta AI, free assistant built into Facebook, Instagram, WhatsApp | Grok by xAI, real-time X and web search built into chat | Perplexity AI, cited-answer research engine | GLM AI, Zhipu’s free chat backed by open models
