The short answer
Kimi vs DeepSeek: pick Kimi, Moonshot AI's assistant on Kimi K3.1 Pro, for multi-step work such as Deep Research reports, slides, documents, spreadsheets and coding with Kimi Code on paid plans. Pick DeepSeek for free reasoning and code: DeepSeek-V4.1-Flash-1017 by default, DeepSeek-V4-Pro in Expert Mode, no paid tier, an in-app training switch and MIT-licensed weights. Both store your data in mainland China, so keep confidential work out of both.
| Kimi | DeepSeek | |
|---|---|---|
| Made by | Moonshot AI, Beijing, China | Hangzhou DeepSeek Artificial Intelligence Co., Ltd., China |
| Chat models | Kimi K3.1 Pro and K3.1 Pro Highspeed | DeepSeek-V4.1-Flash-1017, DeepSeek-V4-Pro in Expert Mode |
| Free plan | Yes; limits not published; Kimi Code paid only | Yes, the whole chat; no paid consumer plan |
| Team plans | Kimi Business and Kimi Enterprise | None for the chat |
| Context window | 1M tokens | 1M tokens |
| Agents | Deep Research, slides, docs, sheets, websites, Agent Swarm, Kimi Code | None; chat with web search |
| Connectors | Gmail, Google Calendar, Google Drive, Notion, Obsidian, GitHub and more | Few |
| Where your data is stored | Mainland China | People's Republic of China |
| Stop training | Contact customer service (no in-app switch) | Switch off Improve the model for everyone |
| Open weights | Kimi K3 under the Kimi K3 License; K3.1 Pro promised by 4 Nov 2026 | V4-Pro and V4.1-Flash-1017 under MIT |
Kimi vs DeepSeek: the verdict
Kimi and DeepSeek both come from Chinese AI labs, both run on very large open-weight models with a 1-million-token context window, and both store your data in mainland China. Past that, they are built for different jobs. Kimi, from Moonshot AI in Beijing, is an agent workspace: it runs on Kimi K3.1 Pro and is organised around tasks such as deep research, slides, documents, spreadsheets, websites and code, with connectors to Gmail, Google Drive and Notion and a desktop app called Kimi Work. DeepSeek, from Hangzhou DeepSeek Artificial Intelligence Co., Ltd., is a lean chat with no paid tier: DeepSeek-V4.1-Flash-1017 by default and DeepSeek-V4-Pro in Expert Mode.
In one line: Kimi for getting multi-step work done, DeepSeek for free reasoning, a cheap API and the cleanest open-weight licence.
Research reports, slides, documents and spreadsheets: Kimi wins
This is Kimi's home ground. Its Help Center lists Deep Research, which writes a longer report, and agents that produce slides, documents and spreadsheets, plus Agent mode and Agent Swarm, which runs several agents in parallel on one task. It can also build, deploy and share web pages from a chat. DeepSeek has none of these: it answers in the chat, with web search from its own search engine, and you turn the answer into a document yourself. If the output you need is a file rather than a reply, start with Kimi.
Coding: Kimi wins for a full coding agent, DeepSeek for free help
Kimi has a dedicated coding product, Kimi Code, with a desktop app, a command-line tool and a VS Code extension, and connectors for GitHub, Cloudflare, Vercel and Supabase. The catch is that Kimi Code is only on paid plans. Moonshot reports that Kimi K3.1 Pro scored 65.7% on Terminal-Bench 4.0 against 66.4% for Claude Opus 5.5, but that is Moonshot's own figure, run with its own setup, not an independent test.
DeepSeek is the better choice if you will not pay. The free chat explains errors, writes scripts and reasons through code, and Expert Mode gives you the larger V4-Pro. For developers, DeepSeek's API works with the OpenAI and Anthropic formats and with Codex and Claude Code. The winner for serious, multi-file coding work is Kimi, because it gives you a coding agent rather than a chat window.
Everyday questions and free use: DeepSeek wins
DeepSeek publishes no paid consumer plan at all, so the free chat is the whole product, Expert Mode included. Kimi has a free tier, but its free limits are not stated on a page we could read, and some of its best features (Kimi Code, more usage) sit on paid membership plans. For someone who wants a capable assistant for quick questions and drafts without ever paying, DeepSeek is the more generous deal. Neither publishes fixed usage limits for its free chat.
Work apps and connectors: Kimi wins
Kimi connects to Gmail, Google Calendar, Google Drive, Notion and Obsidian, supports scheduled tasks, memory and installable skills, and has a browser extension and the Kimi Work desktop app for Windows and Apple silicon Macs. DeepSeek's chat has few connectors or work-app integrations. Be careful with this advantage, though: connecting your work mail and files sends that content to a service whose privacy policy stores data in mainland China, which is exactly the material most companies want to keep out.
Availability outside China and sign-up: a tie
Both run in a browser and on iOS and Android outside China. Both need an account before you can chat, and neither documents its sign-in options on an official page we could open, so try the sign-up screen yourself before you roll either out to a team.
One caution for public-sector work: a US Department of Homeland Security order, published in the Federal Register on 20 October 2026, requires federal civilian agencies to remove DeepSeek technology. It does not ban private use, but check your client contracts if you work with US government bodies.
Data location and privacy: DeepSeek wins, narrowly
Both store personal data in mainland China, so neither belongs anywhere near client data, contracts or personal information about other people. The difference is how easily you can stop training:
- DeepSeek: the privacy policy says data is collected, processed and stored in the People's Republic of China, and its Terms of Use are governed by mainland Chinese law. You can switch off Improve the model for everyone in the app's settings (the setting is named in DeepSeek's Terms of Use), or write to privacy@deepseek.com.
- Kimi: the privacy policy (updated 24 August 2026, effective 31 August 2026) names Beijing Moonshot Technology Co., Ltd. as the operator, stores personal information in mainland China, and says 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 Kimi customer service using the details in section 11 of the policy, and training use stops after identity verification. Memory can be switched off and managed on the settings page.
DeepSeek wins because a settings switch is easier and more reliable than a support request. On the other hand, Kimi offers Kimi Business and Kimi Enterprise plans with member management and admin control over models, which DeepSeek's chat does not have at all.
Open weights and self-hosting: DeepSeek wins
Self-hosting an open-weight model is the only way to use either with sensitive data, because nothing leaves your own servers. DeepSeek wins on licence and timing. DeepSeek-V4-Pro (1.6 trillion parameters, 49 billion active) and DeepSeek-V4.1-Flash-1017, the model behind the chat today, are on Hugging Face under the MIT licence.
Moonshot's latest published weights are Kimi K3: 2.8 trillion parameters with 104 billion active, a 1,048,576-token context, released under the Kimi K3 License. That licence uses MIT-style wording but adds conditions: a separate agreement with Moonshot for Model-as-a-Service businesses with more than US$20 million revenue in 12 months, and a visible "Kimi K3" credit for products with more than 100 million monthly users or US$20 million monthly revenue. Moonshot says the full Kimi K3.1 Pro weights will be released by 4 November 2026; on 1 November its announcement still described them as coming. For most companies both licences are workable, but MIT is simpler for your legal team. Our guide to open-source ChatGPT alternatives explains how to run models like these.
Price: DeepSeek wins
DeepSeek's chat is free and its API is pay as you go, with a public pricing page and off-peak usage at half price since 16 August 2026. Kimi has a free tier and paid membership plans (see the Kimi pricing page), and its API is billed separately on the Kimi Open Platform. If you build on older Kimi models, note that K2 series models closed to newly registered API users on 31 October 2026 and will all be retired by the end of December 2026. On cost alone, DeepSeek is the cheaper and more predictable choice.
Same task, both tools
These come from the features each maker documents, not from tests we ran.
Build a competitor overview deck from public sources
Kimi: run Deep Research on the competitors, then ask the slides agent to turn the report into a deck. One workspace, two steps, a file at the end. DeepSeek: web search gives you a written overview, and you build the slides yourself. Edge: Kimi.
Clean up a spreadsheet of public data
Kimi: the sheets agent can produce a spreadsheet as the output. DeepSeek: paste the data and ask for formulas and a cleaned table in the chat, then copy it back. Edge: Kimi, for getting a file back.
Work through a tricky calculation
Kimi: K3.1 Pro can reason through it. DeepSeek: turn on Expert Mode and ask for each step, with Expert Mode included in the free chat. Edge: DeepSeek, because it is free and the reasoning model is the product's focus.
Refactor a codebase
Kimi: Kimi Code (paid plans) works as an agent from the command line or VS Code. DeepSeek: point Claude Code or Codex at the DeepSeek API for a low-cost engine, or paste code into the chat. Edge: Kimi for a ready-made agent, DeepSeek if you already use Claude Code or Codex. In both cases, sending company code to a service in China is a decision for your security team.
Who should pick Kimi, and who should pick DeepSeek
Pick Kimi if you want an assistant that finishes multi-step jobs (research reports, decks, documents, spreadsheets, small websites), you will pay for Kimi Code, or your team wants a Business or Enterprise plan with member management. Read our Kimi guide first.
Pick DeepSeek if you want a strong reasoning and coding model that is completely free, a cheap API for your developer tools, a training switch you can flip yourself, or MIT-licensed weights to self-host. Read our DeepSeek guide first.
Pick neither for confidential work. Both keep data in mainland China. For work that carries your company's name, compare assistants with business plans that exclude training by default in the best ChatGPT alternatives, see Qwen vs DeepSeek for another Chinese option, or use our comparison tool. The free lessons at /learn?tool=kimi help you practise.
Prompts to copy
Competitor research to deck (Kimi)
Use Deep Research. Research [competitor 1], [competitor 2] and [competitor 3] in the [market] market using public sources only. Cover pricing model, target customer, recent product launches and how each describes itself. 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 research report and a draft deck in one run
Free reasoning check (DeepSeek)
Turn on Expert Mode. Here is a problem with no confidential data: [describe it]. Before answering, list the assumptions you are making. Then solve it step by step, check your own answer a second way, and tell me where a small change in an assumption would change the result.
You get: A worked answer with assumptions and a self-check you can verify
Explain an error without sharing secrets (either)
I have removed names, keys and internal URLs from this error and code: [paste]. Explain in plain English what is going wrong, give the most likely cause and two less likely ones, and show the smallest change that would fix the most likely cause.
You get: A diagnosis and minimal fix without exposing company details