Kimi: Long-Context AI Workspace for Research, Files, Code and Agents

Moonshot AI released Kimi K3 on July 16, 2026, then published the full model weights and technical materials later in the month. The release matters because K3 is not confined to a developer endpoint. It is available across Kimi.com, Kimi Work, Kimi Code, Kimi Agent, Agent Swarm, and the Kimi API, giving the model a direct route into research, office documents, spreadsheets, presentations, websites, local files, browser work, and software development.

K3 is a 2.8-trillion-parameter mixture-of-experts model with 104 billion parameters activated per token. Moonshot says it uses Kimi Delta Attention, Attention Residuals, native visual understanding, and a context window of up to 1,048,576 tokens. In practical terms, the model can accept text, images, video, large documents, and code context within the same workflow. It also always reasons, with low, high, and max reasoning-effort settings available through the API.

The access model needs careful reading. K2.6 chat remains available without consuming membership credits, while K3 and most agentic products draw from Kimi's shared credit pool. Current individual plans are listed at $19, $39, $99, and $199 per month, with lower effective monthly prices for annual billing. Higher plans provide more Agent credits, Swarm uses, Kimi Code capacity, concurrency, and access to K3's extra-long chat. API customers pay separately. K3 currently costs $0.30 per million cached input tokens, $3 per million uncached input tokens, and $15 per million output tokens.

The open-weight release is meaningful, but it should not be confused with easy local use. A 2.8T model requires serious inference infrastructure, and the weights use the Kimi K3 License rather than a familiar permissive software licence. Most individual users will access K3 through Kimi or a hosted provider rather than run the complete model on a laptop.

Moonshot's benchmark results are vendor-reported and should be treated as evidence to investigate, not a substitute for real workflow testing. Its own launch article says overall performance still trails the strongest proprietary models it evaluated, even while K3 performs competitively in several coding, research, office, and agentic tasks. The more useful question for buyers is whether Kimi's connected workflow produces a better finished result for their actual files, sources, and constraints.

WhatAI's take: K3 makes Kimi a serious alternative for people who want one system to move from information gathering to a tangible deliverable. Start with a bounded task, record the sources, credit use, time, and corrections, then compare the same brief in ChatGPT, Claude, Gemini, or Perplexity. The community will learn more from those repeatable results than from headline parameter counts.

WhatAI Decision Box

Kimi is Moonshot AI's connected workspace for chat, web research, office documents, spreadsheets, slides, websites, coding, browser work, and agent execution. Its current flagship, Kimi K3, combines native visual understanding with a context window of up to one million tokens. The product is available through the web, mobile apps, Kimi Work, Kimi Code, WebBridge, Kimi Claw, and an API. This makes Kimi more useful to compare as a working environment than as a single chatbot.

Kimi's clearest advantage is the number of real outputs it can carry from one request. A user can research a topic, analyze source files, run Python, create a document or workbook, generate slides, build a website, or move into a coding workflow without assembling a separate tool for every stage. Agent Swarm adds parallel execution for broad tasks, while Kimi Work and WebBridge extend the system into local files and signed-in browser sessions. The tradeoff is a more complicated product and credit model. K3, Swarm, Code, Work, research, slides, websites, and other membership features share credits, and autonomous work still needs human review.

Kimi is moving quickly from a long-context chatbot into a broad agentic workspace.

Join the conversation below to share which model and mode you used, what files or sources you gave it, how many credits the task consumed, what the final output looked like, and where human correction was still required. Real workflow reports will help the community understand when Kimi is a better choice than ChatGPT, Claude, Gemini, Perplexity, or a specialist tool.

Use Cases

Pricing varies by plan and region — see current pricing.

Plan features change — last updated: 2026-07-29.

Kimi Pros & Cons

How to Get Results with Kimi: Step-by-Step Workflow

1. Choose the final deliverable

   Define whether you need an answer, cited report, Word file, spreadsheet, slides, website, code change, or browser task before selecting a Kimi mode.

2. Select the lightest suitable model

   Use K2.6 for quick chat and ordinary questions, K3 for demanding reasoning and multimodal work, and K3 Swarm only when parallel research or batch execution creates real value.

3. Prepare trusted context

   Upload the relevant files, provide direct source URLs, state the audience and date range, and separate authoritative evidence from background material.

4. Write constraints and approval boundaries

   Specify required formats, prohibited actions, budget or credit limits, privacy boundaries, citation rules, and every action that needs human approval.

5. Run a bounded first pass

   Ask for an outline, plan, sample, or read-only diagnosis before authorizing a long Agent, Swarm, coding, local-file, or browser workflow.

6. Inspect evidence and intermediate work

   Review sources, formulas, code changes, screenshots, task logs, and assumptions while the cost of correction is still low.

7. Export and verify the deliverable

   Open generated Office files, test websites and code, recalculate spreadsheets, follow citations, and check that no browser or local action exceeded the stated scope.

8. Record the real workflow result

   Save the model, mode, plan, credit use, time, prompt, corrections, and final outcome, then share the useful evidence with the WhatAI community.

Kimi Gotchas and Limits to Know Before You Start

• K2.6 chat can be free while K3 and advanced agent features consume credits, so confirm the selected model before starting.
• Membership features share one credit pool. Heavy use of Deep Research, Slides, Code, Work, Claw, websites, K3, or Swarm reduces capacity available elsewhere.
• Agent credit figures are approximate because consumption depends on tokens, complexity, duration, and the feature used.
• Kimi Code has separate five-hour and weekly rate limits in addition to the broader membership credit pool.
• The one-million-token context window is available only through eligible K3 access and does not guarantee perfect retrieval from every part of the context.
• Agent Swarm is a beta feature with plan-based usage and concurrency limits. More subagents can increase duplicated work and review burden.
• WebBridge can operate through existing browser sessions. Use narrow tasks, avoid exposing unrelated tabs, and keep approvals around purchases, publishing, deletion, and account changes.
• Kimi Work can access local folders and run code only within approved scope. Review paths, commands, credentials, and outputs before authorizing changes.
• Kimi Websites does not currently support third-party payment integration, third-party OAuth, or complex external SaaS API integrations.
• Standard cloud Agent cannot directly access local files or an enterprise intranet. Kimi Work or another approved local route is required.
• Generated research can contain weak citations or factual errors. Generated spreadsheets can contain incorrect formulas, and generated code can introduce security or reliability problems.
• K3's full weights are enormous. Open-weight availability does not make full local inference practical on a normal laptop.
• K3 uses the Kimi K3 License. Review its terms instead of assuming the permissions of Apache, MIT, or another standard open-source licence.
• Kimi's many surfaces and rapid releases can make documentation, plan names, credits, and feature availability change quickly.
• Kimi Business makes a specific no-training statement for enterprise data. Do not automatically extend that promise to every consumer product surface or external integration.

Which Kimi Feature Fits Your Use Case

Starter Prompts for Kimi

What is Kimi?

Kimi is an AI workspace from Moonshot AI. It combines chat, web research, multimodal file understanding, documents, spreadsheets, slides, websites, coding, desktop workflows, browser automation, agents, and API access.

Is Kimi free?

Kimi has a free tier. Its help center says K2.6 chat does not consume membership credits. K3, K3 Swarm, Agent, Kimi Code, Kimi Work, Deep Research, Slides, websites, and other advanced features use shared credits or require paid access for higher usage.

How much does Kimi cost?

Individual monthly plans are listed at $19, $39, $99, and $199, with lower effective monthly prices on annual billing. A free tier is available. Kimi Business costs $599 per seat per year with a five-seat minimum, and API use is billed separately by tokens.

What is Kimi K3?

Kimi K3 is Moonshot AI's flagship native multimodal model. It has 2.8 trillion total parameters, 104 billion activated parameters, visual understanding, and a context window of up to one million tokens.

Can Kimi read images, video, PDFs, Word files, and spreadsheets?

Yes. Kimi K3 supports text, image, and video input, and Kimi's workspace includes dedicated document and spreadsheet workflows. Results still need review, especially when the source is visually complex or the task requires exact calculations.

What is Kimi Agent Swarm?

Agent Swarm is Kimi's parallel execution mode for large research, writing, and batch tasks. Moonshot says its orchestrator can coordinate up to 300 subagents and thousands of tool calls, but membership plans limit available uses and concurrent subtasks.

Does Kimi create editable Office files?

Yes. Kimi Docs, Sheets, and Slides can create and revise files including .docx, .xlsx, PDF, and presentation outputs. Users should verify formulas, citations, formatting, and business conclusions before relying on the files.

Can Kimi build and publish websites?

Yes. Kimi Websites can generate and deploy supported frontend and full-stack projects. Current official limitations include third-party payments, third-party OAuth, and complex external SaaS API integrations.

What is Kimi Work?

Kimi Work is a desktop agent for Windows and Apple-silicon macOS. It can mount local folders, run Python, use WebBridge, and execute scheduled workflows. Its Ask before acting safeguard requests authorization before local modifications or code execution.

What is Kimi WebBridge?

WebBridge is a Chrome and Edge extension plus a local bridge service that lets supported agents navigate, click, fill forms, take screenshots, and extract information through an existing browser session. It should be given narrow instructions because signed-in sessions can expose powerful account actions.

Does Kimi support developers?

Yes. Kimi offers Kimi Code for terminal and IDE workflows, MCP support in its CLI, an OpenAI-compatible API, an Anthropic-compatible API, and open K3 model weights for supported deployment environments.

Is Kimi K3 open source?

Moonshot has released Kimi K3's full model weights and code resources under the Kimi K3 License. Treat it as an open-weight release and review the licence before commercial use, redistribution, or modification.

Does Kimi have unlimited use?

No. Paid membership features share one credit pool, Kimi Code also has separate five-hour and weekly limits, Agent Swarm has plan-based use and concurrency limits, and API usage is billed by tokens.

Is Kimi accurate?

Kimi can search, cite, calculate, and work across large files, but it can still misread evidence, invent details, produce incorrect code, or create flawed formulas. Important outputs need source checks, tests, and human approval.

Related AI Tools & Chatbots Tools

Explore the Network