Kimi Agent Swarm
Agent swarm in the Kimi AI assistant from Moonshot AI
Kimi Agent Swarm is a feature of Moonshot AI's Kimi assistant for large, parallelizable tasks. An orchestrator sets the strategy and assigns tasks to up to 300 subagents that work simultaneously without predefined roles; a single task can involve more than 4,000 tool calls. The feature currently runs on the Kimi K3 model (K3 Swarm) and produces code projects, data analyses and Word, PDF and PPT documents, among other outputs. Agent Swarm is labeled beta and available in the paid Moderato, Allegretto and Allegro memberships; swarm tasks consume several times the credits of a regular agent task.
Good fit if
Focus: Users who want very broad research, bulk downloads or long documents produced in a single run
- your project is about Large-scale information retrieval, Reading and analyzing 100+ documents or Long structured texts
- people without programming skills should work with it (5/5 for use without developers)
- your team is just getting started (skill level: Beginner)
- you want to get started quickly (learning curve: flat)
Key figures
- Usable without developers
- 5/5
- Scalability
- 4/5
- Integrations
- 2/5
- Versioning
- 1/5
- Observability
- 2/5
Editorial rating on a scale from 1 to 5. Methodology
Fact sheet
- Category
- AI assistants
- Type
- Agent swarm in the Kimi AI assistant from Moonshot AI
- Licence
- Proprietär
- Pricing
- FreemiumKimi can be used for free; Agent Swarm requires a paid membership. Monthly prices per the Kimi help center: Andante ¥49 (without Agent Swarm), Moderato ¥99, Allegretto ¥199, Allegro ¥699; annual plans are cheaper. Concurrent swarm subtasks are 2, 4 or 8 depending on the tier.
- Deployment
- Cloud, Mobile
- Models
- Moonshot AI (Kimi)
- RAG
- Yes
- Multi-Agent
- Yes
- SSO, roles (RBAC)
- SSO no, RBAC no
Key features
- Agent Swarm with an orchestrator and up to 300 subagents
- More than 4,000 tool calls per task
- Output as code, tables, documents and presentations
- Web and Kimi app
- Credit-based billing
Typical use cases
- Large-scale information retrieval
- Reading and analyzing 100+ documents
- Long structured texts
- Front-end development and refactoring
- Office automation with documents and presentations
Market notes
Changes to Kimi Agent Swarm by email
We let you know when the status, pricing or license of Kimi Agent Swarm changes, together with the weekly market notes.
Implementation
Get Kimi Agent Swarm implemented
Free for you. Partners only pay for requests they accept.
Alternatives
Similar systems
Similar in category, features or use cases.
- AI assistantsManusManus is a general-purpose AI agent that runs research, data and creation tasks in its own cloud environment; the company has been part of Meta since late 2025 and continues to operate its subscription offering. Wide Research splits tasks with many similar items into independent subtasks, distributes them to subagents working in parallel with their own context window and merges the results in the main agent. Each subagent is a fully capable Manus instance. The feature starts automatically when Manus considers a task parallelizable and is available to paying users. Plans are Free, Pro and Team with a credit model.View profile
- AI assistantsPerplexity ComputerPerplexity Computer is an agent from Perplexity, introduced in February 2026, that completes multi-step tasks in an isolated compute environment with a file system, browser and tool integrations. It breaks a task into subtasks and creates sub-agents that work asynchronously; models from different providers are used depending on the subtask, and users can also pick models for subagents themselves. Personal Computer extends this on the Mac with local files and local computer use, and there is also a Slack integration. Computer is available in Pro, Max, Enterprise Pro and Enterprise Max and is billed through credits.View profile
- AI assistantsChatGPT (Work und Codex)ChatGPT is OpenAI's AI assistant; ChatGPT Work is an agent that completes tasks across apps and files, and Codex is the coding agent. ChatGPT Work and Codex can run subagent workflows by spawning specialized agents in parallel, for example on a direct request in chat or as instructed by AGENTS.md or skills. Custom agents are defined as TOML files with a name, description and instructions; built-in types are default, worker and explorer. ChatGPT agent mode itself, by contrast, is described by OpenAI as a single unified agentic system. Plans are Free, Go, Plus, Pro, Business, and Enterprise and Edu.View profile
- AI assistantsClaudeClaude is Anthropic's AI assistant for web, desktop and mobile, complemented by the coding tool Claude Code. The Research feature follows an orchestrator-worker pattern: a lead agent plans the research and spawns subagents that explore different aspects in parallel. In Claude Code the main session delegates tasks to subagents with their own context window, tools and permissions; the experimental agent teams coordinate several Claude Code instances through a shared task list and direct messaging. Claude targets individuals, teams and enterprises; there is a free plan plus Pro, Max, Team and Enterprise.View profile