The short answer
Kimi vs Claude: pick Claude, Anthropic's assistant on Claude Opus 5.5, for most work. It wins writing, coding, files and privacy, and Team and Enterprise content is not used for training by default. Pick Kimi, Moonshot AI's assistant on Kimi K3.1 Pro, for public research, 3D Make and open weights: Kimi K3.1 Pro weights went up on Hugging Face in early November 2026 under a licence with conditions for very large businesses. Kimi stores personal information in mainland China, so keep confidential work out of it.
| Kimi | Claude | |
|---|---|---|
| Made by | Moonshot AI, Beijing, China | Anthropic, United States |
| Top models | Kimi K3.1 Pro and K3.1 Pro Highspeed | Claude Opus 5.5 and Claude Fable 5.1 (paid plans) |
| Free plan | Yes; limits not published; Kimi Code sold through paid plans | Claude Sonnet 5.5 and Haiku 5.5, limits every 5 hours and every week |
| Team plans | Kimi Business and Kimi Enterprise | Team and Enterprise |
| Context window | 1M tokens (Kimi K3.1 Pro) | 1M tokens in the API for the 5.5 models |
| Coding agent | Kimi Code: desktop, CLI, VS Code | Claude Code from Pro upwards |
| Files and agents | Deep Research, Slides, Docs, Sheets, websites, Agent Swarm, 3D Make | Cowork, Research, Docs, Slides, Design, Dashboards (beta) |
| Where your data is stored | Mainland China | US or other countries outside the EEA and UK; EU controller in Ireland |
| Stop training | Contact customer service (no in-app switch) | Help improve our AI models toggle; off by default on Team and Enterprise |
| Open weights | Kimi K3 and K3.1 Pro on Hugging Face under Kimi licences with conditions for very large businesses | None |
Kimi vs Claude: the verdict
Pick Claude for most work. Anthropic's assistant wins writing, coding, documents and decks, work apps and privacy, because it finishes whole tasks, hands back files you can open in Word, PowerPoint or Google Docs, and offers Team and Enterprise plans where your content is not used for training by default. Pick Kimi, from Moonshot AI in Beijing, when you want a capable agent at a lower commitment, 3D models and interactive pages, or open weights you can run on your own servers. Kimi's privacy policy stores personal information in mainland China, which rules it out for client and company data.
Here is where each one stands on 4 November 2026. Kimi runs on Kimi K3.1 Pro, which Moonshot rolled out in late October across Kimi.com, the iOS, Android and HarmonyOS apps, the Kimi Work desktop app (version 3.1.0 or later), Kimi Code and the Kimi API, with a faster companion, Kimi K3.1 Pro Highspeed, that Moonshot says runs 2 to 3 times faster. Claude runs on Anthropic's 5.5 models: Claude Opus 5.5 (released 22 September 2026) on Pro and above, and Claude Sonnet 5.5 (28 September) and Claude Haiku 5.5 (7 October) on every plan including Free. Since September 2026, chat and Claude Cowork are one experience, so any Claude conversation can take on a task and return editable files.
In one line: Claude for work that carries your name, Kimi for public material, experiments and self-hosting.
Writing and editing: Claude wins
Careful writing and editing is what Claude is best known for, and Anthropic says Opus 5.5 writes more clearly and leads with the most important information. The bigger difference is what happens after the draft. Claude Docs, Slides and Design are on every plan, including Free, so a draft becomes a document you export to Word, PDF, Markdown or Google Docs, or a deck you export to PowerPoint, PDF or Google Slides. Memory works across chat and cloud Cowork tasks, and you can edit what it remembers in Settings > Your Claude (Settings > Memory in the older layout), which keeps your tone and context consistent from one piece to the next.
Kimi writes well too, and its Docs agent produces a document as the output. But most work writing contains names, figures and opinions about real people, and Kimi's policy sends that to servers in mainland China. For a newsletter about public topics Kimi is fine; for a client report, use Claude.
Coding: Claude wins, narrowly
This is the closest contest on the page. Anthropic reports 66.4% for Claude Opus 5.5 on Terminal-Bench 4.0, a test of coding agents working in a terminal. Moonshot reports 65.7% for Kimi K3.1 Pro on the same test, run with the Claude Code harness, and describes that as narrowly behind Opus 5.5. Both numbers come from the makers themselves, not from an independent test, so read them as "roughly level" rather than as a ranking.
Both have a dedicated coding agent. Claude Code comes with Pro and every plan above it, and Anthropic says Opus 5.5 handles long coding jobs well. Kimi Code runs on Kimi K3.1 Pro as a desktop app, a command-line tool and a VS Code extension, and you switch to K3.1 Pro with the /model command; it is sold through paid Kimi Code plans. Claude wins for professional teams because the code you send stays with a US provider, Team and Enterprise content is not used for training by default, and Enterprise adds audit logs and custom data retention. Kimi is a credible choice for side projects, open-source work and anything you are happy to send to a provider in China.
Research reports, decks and spreadsheets: Claude wins for work, Kimi for breadth on public topics
Both assistants are built to hand back files rather than replies. Kimi's Help Center lists Deep Research, which writes a longer report, agents that produce slides, documents and spreadsheets, Agent mode and Agent Swarm, which runs several agents in parallel on one task, plus scheduled tasks, memory and installable skills. Moonshot also positions Kimi K3.1 Pro for end-to-end financial workflows, which makes it worth a try for modelling on public market data.
Claude covers the same jobs with a different shape. Research, its multi-step report mode, is on Pro and above. Docs, Slides and Design are on every plan. Claude Dashboards (beta) on paid plans build charts from data platforms or connected apps such as Salesforce, and every chart shows its query and last refresh, which makes the numbers checkable. Claude wins for anything you will send to a client or a manager, because the files export straight to the formats your company already uses and the data stays out of China. For a quick deck on a public topic, Kimi is a strong runner-up.
Long documents: a tie
Moonshot gives Kimi K3.1 Pro a 1-million-token context window. Anthropic's developer documentation gives Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5 a 1-million-token context window in the API, with up to 128k output tokens per request. Anthropic does not publish the context size inside the Claude apps by plan, so do not assume the app matches the API. In practice, both can take a long contract, a large codebase or a stack of reports in one go. Claude's Projects (up to five on Free, unlimited on Pro) are the better home for material you come back to every week.
Work apps and connectors: Claude wins
Claude connects to Gmail, Google Calendar and Google Drive for all users, and asks before it sends, replies to or forwards an email. Pro and above add the Claude for Microsoft 365 add-ins (the Microsoft 365 connector itself works on every plan, Free included) and Claude in Chrome, and Salesforce in Claude is in beta. Team and Enterprise owners decide which connectors their organisation can use.
Kimi also connects to Gmail, Google Calendar, Google Drive, Notion and Obsidian, plus GitHub, Cloudflare, Vercel and Supabase for coding, according to trade publication VKTR, and has a browser extension and the Kimi Work desktop app for Windows and Apple silicon Macs. The connector list is respectable, but connecting your work mail and files to Kimi sends that content to a service that stores personal information in mainland China. Claude wins on breadth and on whether your IT team will allow it.
Privacy and data location: Claude wins
- Kimi. The privacy policy (updated 24 August 2026, effective 31 August 2026) names Moonshot AI's Beijing company as the operator and says personal information is stored within the People's Republic of China. Inputs and outputs may be used to optimise models after encryption and de-identification. There is no in-app switch: to opt out you contact customer service through the channels in section 11 of the policy (the in-product feedback feature or the policy email address), and training use stops after identity verification. Memory can be switched off and managed on the product settings page. For developers, Moonshot says Zero Data Retention is available for Kimi API customers.
- Claude. Anthropic's privacy policy (effective 10 September 2026) names Anthropic Ireland, Limited as the controller for people in the EEA, UK and Switzerland, and Anthropic PBC for everyone else. Data goes to servers in the US or other countries outside the EEA and UK, with standard contractual clauses or adequacy decisions for transfers out of Europe. On Free, Pro, Max and Ultra your chats can be used for training unless you switch off Help improve our AI models in Settings > Privacy (claude.ai/settings/data-privacy-controls); the same toggle is under your name, then Settings, then Privacy on mobile. Turning it off covers new and stored chats from then on, not training already done, and chats flagged for safety review may still be used. Team and Enterprise content is not used for training by default, and incognito chats are never used for training.
Claude wins because the training switch is a toggle you control, work plans exclude training by default, and European users get an EU controller. Kimi's only route to full control is the one Claude does not offer at all: running the open weights yourself.
3D models and interactive pages: Kimi wins
Kimi has a feature Claude does not: 3D Make, live on kimi.com and in the Kimi app since the K3.1 Pro launch, lets you preview and edit 3D models online, download them for 3D printing, or export them as videos or Live Photos. Kimi can also build, deploy and share web pages from a chat. Claude Design produces mockups and designs you export as a zip, PDF, PowerPoint or HTML, which suits product and marketing teams, but it does not make printable 3D models. If you work in product design, education or hardware prototyping, Kimi is worth an afternoon.
Open weights and self-hosting: Kimi wins
This one is not close. Anthropic publishes no model weights: its Hugging Face page lists datasets and no models, so you always use Claude through Anthropic's apps, its API or the cloud platforms it partners with.
Moonshot publishes its models. Kimi K3 (2.8 trillion parameters, 104 billion active, a 1,048,576-token context) is on Hugging Face under the Kimi K3 License. That licence uses MIT-style permissions with conditions: a Model-as-a-Service business with more than US$20 million revenue over 12 months needs a separate agreement with Moonshot, and products with more than 100 million monthly users or more than US$20 million monthly revenue must show "Kimi K3" in the interface. Internal use is exempt from both. The full Kimi K3.1 Pro weights followed: they are on Hugging Face (published by 5 November 2026, a day after we last found nothing) under a new Kimi K3.1-Pro License with the same two conditions and the same exemption for internal use, except that the required credit reads "Kimi K3.1-Pro". Moonshot's model card lists 2.8 trillion parameters, 104 billion active and a 1,048,576-token context. Self-hosting a model this size needs serious server hardware, so most teams will use the Kimi API instead.
Free plans and price: Claude wins on clarity
Claude's pricing page spells out what Free includes: Claude Sonnet 5.5 and Claude Haiku 5.5, limited usage every five hours and every week, web search, file creation, code execution, memory, connectors, Docs, Slides and Design, and up to five Projects. Opus 5.5, Claude Code, Research and Claude in Chrome need Pro or above. Kimi can be used free on the web and in its apps, but Moonshot does not publish the free limits on a page we could read, and Kimi Code is sold through paid Kimi Code plans. Both charge for their APIs separately. Claude wins because you know what you are getting before you start. Check current prices on Claude's pricing page and the Kimi pricing page.
Same task, both tools
These come from the features each maker documents, not from tests we ran.
Turn meeting notes into a client proposal
Claude: paste the notes and ask for a proposal; it can come back as a document you export to Word or Google Docs, and on Team or Enterprise the content is not used for training by default. Kimi: the Docs agent can draft it, but you would need to remove every client detail first because the data is stored in mainland China. Edge: Claude.
Build a competitor deck from public sources
Kimi: run Deep Research, then ask the Slides agent to turn the report into a deck, all in one workspace. Claude: Research on Pro, then Claude Slides, exported to PowerPoint or Google Slides. Edge: Claude if the deck goes to a client, Kimi if you are on a free plan and the topic is fully public.
Fix a bug in a work repository
Claude: Claude Code on Pro or above works in the repository and runs the code. Kimi: Kimi Code does the same from the command line or VS Code on a paid plan, but sending company code to a provider in China is a decision for your security team. Edge: Claude.
Make a printable 3D model of a product idea
Kimi: describe it to 3D Make, edit the preview and download it for 3D printing. Claude: Claude Design gives you a 2D mockup, not a printable model. Edge: Kimi.
Who should pick Claude, and who should pick Kimi
Pick Claude if you write for clients or managers, want finished documents and decks in the formats your company uses, code professionally, or need admin controls and training off by default. Teams should start on Team. Our guide to using Claude explains every plan, and the free lessons at /learn?tool=claude help you practise.
Pick Kimi if you want a strong agent for public research, slides and spreadsheets, you are curious about 3D Make, or your company wants open weights to run on its own servers, where nothing reaches Moonshot at all. Read our Kimi guide first, and see Kimi vs DeepSeek if you are choosing between Chinese assistants.
Not sure? Use Claude for anything with your company's name on it and Kimi for learning and public material. For other options see Claude vs DeepSeek, the best Claude alternatives and our comparison tool.
Prompts to copy
Same three tasks, scored side by side (either)
I am comparing AI assistants for my job as a [role]. Here is a real task from last week with all names and confidential details removed: [paste task and material]. Do the task. Then list the three assumptions you made, the parts of your answer I should check before using it, and what extra information would have made it better.
You get: A finished answer plus a self-check, so you can score Kimi and Claude on the same work
Turn notes into an exportable proposal (Claude)
Here are my notes from a meeting with [client type]: [paste]. Turn them into a two-page proposal with an overview, the problem, our approach, a timeline and next steps. Keep it plain and specific, flag anything in my notes that is unclear, and give it to me as a document I can export to Word.
You get: An editable proposal and a list of gaps to fill
Public research to deck (Kimi)
Use Deep Research. Research [topic or market] using public sources only. Cover the main players, recent launches and the three trends that matter most for [audience]. Then turn the report into a [8]-slide deck with one finding per slide and a final slide of open questions. Cite a source for every claim.
You get: A sourced report and a draft deck in one run, on material with nothing confidential in it