An AI assistant developed by China-based Moonshot AI. Launched in October 2023 as “Kimi Chat,” it has been known from the start for its strength in handling very long inputs. Today it goes well beyond chat: it bundles agentic task execution, scheduled tasks, real-time web search, coding assistance (“Kimi Code”), and knowledge work support (“Kimi Work”) into a single service. It runs on Moonshot AI’s own Kimi K2 and K3 model families, and because the API is publicly available, you can use it as an app or embed it in your own product.
Key Features
- Very long context handling: Kimi’s starting point and biggest strength. The K3 model available on higher tiers handles inputs on the order of one million tokens, so you can load large documents or source trees at once and ask questions about them
- Agentic task execution: Multi-step work such as “research it, summarize it, and turn it into a document” can be handed off after a single instruction. Depending on the plan, you can run several tasks concurrently or fan work out across subagents
- Scheduled tasks and widgets: You can set recurring jobs such as “collect news matching these criteria every morning,” and keep the results visible as widgets on a dashboard
- Real-time search and multimodal input: Web search is combined with the model so answers reflect current information. It accepts images and files in addition to text
- Kimi Code (coding assistance): Provided as a CLI tool for delegating code generation, refactoring, and test writing. Its cost-conscious design has driven adoption in the developer community
- Kimi Work (knowledge work support): Covers office-oriented work such as research, document preparation, and slide generation. Files can be kept per project so context carries across sessions
- API access: The API is available through Moonshot AI’s platform for integration into your own applications, and the official documentation also covers setting up an MCP (Model Context Protocol) server
Pricing
| Plan | Monthly | Annual | What’s included |
|---|---|---|---|
| Adagio (free) | $0 | ─ | Basic chat, file uploads, web search |
| Moderato | $19 | $180 ($15/mo equivalent) | 60 credits, 2 concurrent tasks, 10 scheduled/widget tasks, 20 projects (20GB storage) |
| Allegretto | $39 | $372 ($31/mo equivalent) | 150 credits, Goal Mode, Kimi Claw, 5,000 professional database calls |
| Allegro | $99 | $948 ($79/mo equivalent) | 360 credits, K3 extra-long chat (up to 1M tokens), 4 concurrent tasks, 8 subagents, 12,000 database calls |
| Vivace | $199 | $1,908 ($159/mo equivalent) | 720 credits, maximum allowances, 25 scheduled tasks, 100 projects, 24,000 database calls |
The API is billed separately from these subscriptions on a per-token basis. Rates differ by model generation (K2 family vs. K3 family), so check the official API documentation for current figures.
Pricing is current as of August 2026. Check the official site for the latest information.
Pros & Cons
✅ Pros
- Consistently strong at long-form processing, which makes it easy to work with large documents or codebases in one pass
- Chat, agents, coding, and knowledge work live in one service, so you don’t have to move between tools by task
- The free plan already covers basic chat, file uploads, and web search, making it easy to try
- Five tiers let you match concurrency and credit volume to what you actually need
- The public API means you can move from app usage to embedding the same models in your own service
⚠️ Cons
- The credit system means heavy agent use quickly pushes you toward higher tiers, and monthly cost is hard to predict
- It is provided by a Chinese company, and some organizations avoid it for business use on data handling or jurisdiction grounds
- Features such as Goal Mode and Kimi Claw are limited to higher tiers and cannot be tried on the free or entry plans
- Japanese-language documentation and support are thinner than the English and Chinese material, so tracking feature details takes extra effort
- The model release cycle is fast, so the models and limits available within a plan change frequently
Comparison with Similar Services
| Criteria | Kimi | ChatGPT | Claude | DeepSeek |
|---|---|---|---|---|
| Provider | Moonshot AI (China) | OpenAI (US) | Anthropic (US) | DeepSeek (China) |
| Strength | Very long context, integrated agents | Breadth of features, image generation | Long context, natural writing | Low cost and reasoning performance |
| Agent features | Built in (concurrency, subagents) | Yes | Yes | Limited |
| Coding assistance | Kimi Code (CLI) | Codex and others | Claude Code | Coding-oriented models |
| Free plan | Yes (Adagio) | Yes | Yes | Yes |
| Paid entry point | From $19/mo | From $20/mo | From $20/mo | Mainly pay-as-you-go API |
Who Is It For
- People who want to load hundreds of pages of material or a large codebase in one go, without splitting it up
- People who want to hand the research-to-document pipeline to an agent and focus on review and finishing
- People who want to automate recurring information gathering or report writing with scheduled tasks
- Developers who want to delegate work from a CLI while keeping coding assistance costs down
- People planning to embed a chat API in their own service who want to try the same models in an app first
Summary
Kimi is an AI assistant that has built outward from long-context strength to bundle agents, scheduled tasks, coding assistance, and knowledge work into a single service. A clear path is to try basic chat and search on the free plan, then pick a paid tier by the credit volume you need once you start using agents seriously. On the other hand, the Chinese provider base and the difficulty of predicting credit-driven costs are both worth checking before rolling it out at work.