Kimi

Kimi is a multimodal generalist AI assistant from Moonshot AI that combines conversational chat with autonomous agent capabilities for large-scale automation. The latest Kimi K2.6 model (released April 2026) ships with 1 trillion total parameters, 32B active per token via Mixture-of-Experts, a 262,144-token context window, and an Agent Swarm system that scales to 300 domain-specialized sub-agents in a single autonomous run. Available as a Progressive Web App on desktop, an Android mobile app, and a public REST API on platform.moonshot.ai.

What does Kimi do?

• •Agent Swarm (K2.6) Orchestrate up to 300 domain-specialized sub-agents executing up to 4,000 coordinated steps in a single autonomous run for massive research, analysis, or deployment projects.
• •Deep Research Mode Conducts multi-source investigations on complex topics, synthesizing information from documents, websites, and datasets into actionable summaries.
• •Design-to-Code Converts screenshots, mockups, or visual designs directly into production-ready front-end HTML, CSS, and JavaScript code, enabling visual prototyping.
• •OK Computer Runs Executes predefined, self-directed workflows where the AI completes a series of steps autonomously based on a single high-level goal you set.
• •Document and Spreadsheet Creation Generates office-style deliverables like presentations, reports, and data sheets from prompts or analyzed content, reducing app switching.
• •Kimi Code CLI Provides a terminal-based command-line interface for builder assistance, allowing code generation and task execution directly from your shell.
• •Public REST API Moonshot AI ships a public Kimi API at platform.moonshot.ai with pay-as-you-go pricing on K2.6: $0.95/1M input tokens, $4.00/1M output tokens, $0.16/1M cache-hit tokens.
• •Browser Extension Integrates Kimi's analysis and chat capabilities directly into your web browser for quick research and content interaction on any page.
• •MCP Support Connects via Model Context Protocol to other tools and data sources, enabling Kimi agents to fetch information and act across your digital workspace.
• •256K Token Context Processes and analyzes extremely long documents, legal contracts, or research papers within a single 262,144-token context window.

What are Kimi's limitations?

• •Registration favors Chinese phone numbers . Signing up for Kimi's full-featured platform typically requires a Chinese mobile number for verification. International users can get access but the setup is clunky and not officially smoothed out.
• •Rate limits are tight for new accounts . Fresh accounts are subject to strict usage limits. Requesting an increase requires submitting a business case in Chinese, making it practically inaccessible for non-Chinese-speaking users who need higher volume quickly.
• •Agentic tasks sometimes end with instructions instead of results . Users report that for some multi-step tasks, Kimi responds by telling you how to complete the task rather than completing it. K2.6 improved long-horizon stability versus K2.5, but the gap with Claude or GPT-5 on autonomous completion is still real.
• •Higher latency outside China . Because Kimi is hosted on Chinese infrastructure, users in Europe and North America experience noticeably slower response times compared to US-hosted alternatives like Claude or ChatGPT.
• •Smaller ecosystem and thinner enterprise support . Kimi has a younger builder community, fewer third-party integrations, and less established enterprise tooling. For business-critical workflows, the support infrastructure is meaningfully thinner than OpenAI or Anthropic.

How much does Kimi cost?

• •Free tier $0/mo. Web chat, basic agent features, browser extension. Word and run quotas apply.
• •Moderato $19/mo. Includes 20 Deep Research runs, 20 OK Computer runs, and 2,048 Kimi Code CLI requests per week.
• •K2.6 API (pay-as-you-go) $0.95/1M input tokens, $4.00/1M output tokens, $0.16/1M cache-hit tokens (cache discount).
• •Volume discounts Custom enterprise pricing available for high-throughput accounts via Moonshot AI sales.