Kimi vs Qwen comes down to the work you need to finish. Kimi is a useful starting point for research, documents, spreadsheets, presentations and longer knowledge-work tasks. Qwen is a useful starting point for broader image, video and audio workflows, a wider choice of downloadable models, and lower-priced text API options. For everyday writing and questions, compare their answers on your own material.
These are product-fit judgments based on documented capabilities. A broader model catalog or a more integrated workflow does not prove that one assistant gives more accurate answers. Identify the product first: consumer chat, a desktop work agent, a coding service, a named API model or a self-hosted checkpoint.
Last verified: October 13, 2026. This is a review of official documentation, pricing, model information and data policies. No matched hands-on benchmark was performed for this article. Research and editorial review were AI-assisted; see our methodology. ChinaAI is independent of Moonshot AI, Kimi, Alibaba and Qwen.
On this page: Quick verdict · Products · Free access and plans · Research and PDFs · Work agents · Coding · Images and media · APIs and costs · Local models · Privacy
Kimi vs Qwen: the quick verdict
| Your priority | Where to start | What should decide the choice |
|---|---|---|
| Everyday chat and writing | Try both | Factual accuracy, instruction following and how much editing the answer needs. |
| Research that becomes a report, spreadsheet or presentation | Kimi | Its integrated research and office workflows are relevant; verify citations and exported files. |
| Local files and desktop work | Compare Kimi Work with QwenWork | Required applications, permissions, operating-system support and the finished deliverable. |
| Repository-level coding | Compare Kimi Code with Qwen Code | The same issue, repository, tests, model access and total cost. |
| Image, video and audio creation | Qwen’s wider ecosystem | Select the exact media model or tool; catalog breadth does not establish output quality. |
| Lower-priced text API processing | Evaluate Qwen3.8 Flash | It has the lowest input/output rates in the example below; check task quality and retries. |
| Deployment across different hardware tiers | Explore Qwen’s model catalog | The exact checkpoint, license, memory requirement, quantization and runtime. |
| Confidential information | Choose an appropriate service contract | Data use, retention, location, permissions and all external tools involved. |
Use the table to choose a first evaluation, then inspect the result. For example, a research workflow should be judged by supported claims and usable files, while a coding workflow should be judged by its patch and tests.
Which products are being compared?
Kimi and Qwen each cover several products. A result from a hosted API does not automatically describe the consumer app, and a local model does not reproduce every tool in a cloud assistant.
| Product layer | Kimi route | Qwen route | What to compare |
|---|---|---|---|
| Consumer assistant | International Kimi or the China service | Qwen consumer assistant | Available modes, files, search and account benefits. |
| Desktop and deliverable workflows | Kimi Work | QwenWork | Local-file access, browser operations, documents and recurring tasks. |
| Coding agent | Kimi Code | Qwen Code | Repository actions, supported clients, permissions and inference allowance. |
| First-party developer API | Kimi API Platform | QwenCloud or an applicable Alibaba Cloud Model Studio route | Exact model, provider, endpoint, region, tools and billing. |
| Downloadable model | Official Kimi K3 checkpoint | Official Qwen checkpoints | License, model size, runtime and hardware. |
Qwen consumer-use scope: the Service Agreement, effective September 10, 2026, limits that consumer service to private, non-commercial use. For commercial work, review the agreement for the exact QwenWork, API, hosted provider or self-hosted model route you intend to use.
Free access and plans: China, international and legacy Kimi memberships
Consumer allowances, work-agent subscriptions, coding benefits and API balances are different products. Kimi publishes separate China and international membership menus. An English-language help page on the China site does not make its yuan prices an international offer.
Current Kimi China plans
The current China plans are Go, Plus, Pro and Max. The following prices are in Chinese yuan (CNY), as listed in the official China pricing documentation.
| Current China plan | Monthly billing | Annual billing, paid upfront | Included Kimi Code quota |
|---|---|---|---|
| Go | ¥49/month | ¥468/year | No |
| Plus | ¥99/month | ¥948/year | Yes |
| Pro | ¥199/month | ¥1,908/year | Yes |
| Max | ¥699/month | ¥6,708/year | Yes |
The annual prices are equivalent to ¥39, ¥79, ¥159 and ¥559 per month, respectively. Those are comparison figures: the annual total is charged upfront, rather than collected as twelve payments at the discounted monthly equivalent.
The China benefits overview includes document work, presentations, Deep Research and website deployment across these plans. Plus adds Kimi Code and Agent Swarm; Pro adds Goal Mode and a dedicated Kimi Claw; Max provides a larger credit allowance and higher Agent-task concurrency. Features remain subject to their usage limits.
What happened to Andante, Moderato, Allegretto and Allegro?
They are legacy China plans. Kimi says existing subscribers can retain their plan names, benefit rules and auto-renewal arrangements, continue renewing, or choose an available upgrade. Moving to a new plan is optional. The legacy names should not be presented as the current menu for new subscriptions.
The change is not just a new label for identical benefits. For example, legacy Andante retains Kimi Code access, while new Go does not include a coding quota. The official Kimi Code guide also distinguishes the limits: new eligible plans retain a rolling five-hour usage window without a weekly limit; legacy plans retain both the rolling five-hour and weekly limits. This does not mean users are allowed only five hours of coding per week.
Eligible features still draw from the shared monthly membership allowance. Both monthly and annual subscriptions refresh their cycle credits monthly, and unused cycle credits do not roll over. Removing the weekly Code limit does not create unlimited usage. Review the benefits shown for your existing plan before switching.
International Kimi plans are a separate menu
The official international pricing page lists Plus at $19, Pro at $39, Max at $99 and Ultra at $199 per month. All four new international plans include Kimi Code. Its legacy names are Moderato, Allegretto, Allegro and Vivace.
Do not treat an international Plus, Pro or Max as a currency conversion of a China plan with the same name. Use the regional benefits, billing cycle and checkout applicable to your account. Availability, taxes, app-store billing and promotions can affect the purchase. Our Kimi pricing guide explains the separate consumer and developer routes.
How does Qwen’s billing compare?
The ordinary Qwen consumer assistant is publicly described as free to use. That does not establish unlimited access, free API calls or inclusion of QwenWork. Kimi also offers a free allowance, with memberships expanding eligible usage.
| QwenWork personal plan | Monthly billing | Annual billing equivalent | Monthly plan credits |
|---|---|---|---|
| Free | $0 | Not applicable | Limited free allowances and promotional credits; check the current offer. |
| Plus | $20/month | $16/month, billed annually | 2,000 |
| Pro | $40/month | $32/month, billed annually | 4,500 |
These are the figures on the QwenWork pricing page, not a universal subscription for every Qwen product. Kimi and QwenWork credits are different accounting units and should not be compared one for one. General Kimi API billing, QwenCloud and Model Studio billing remain separate from the consumer memberships. See our Qwen pricing guide for the product and provider distinctions.
Research, writing, PDFs and long documents
Research and writing
Kimi is a practical first test when research must become a report, document, spreadsheet or presentation in one workflow. Its capability overview and Deep Research documentation describe search, research and deliverable creation. Professional databases and plugins can extend that workflow, subject to access and plan benefits.
Qwen and QwenWork are relevant alternatives when the same project involves more visual material, webpages or a later developer integration. For a straightforward explanation, translation or writing draft, the documented product differences are not enough to name a quality winner.
A useful writing comparison starts with the same audience, source material, required facts and length. Check factual correctness before style, then give both tools the same revision request. Preserve the first answers so that an attractive final draft does not hide how much correction it needed.
For research, open the original source behind each important claim. Check whether it supports the complete sentence, uses the right geography and time period, and preserves qualifications. A long reference list is not evidence that every statement was verified.
PDFs and long-document analysis
Both support document workflows. Kimi combines uploaded-file analysis with Projects and local folders through Kimi Work. Qwen provides consumer document processing, while QwenWork adds workspace and local-file workflows. A large API context window does not establish the file limits or retrieval behavior of the consumer interface.
To compare PDF handling, use one small searchable document with known answers. Ask for a fact, a table value and their page references, plus one fact that is absent. Require “Not stated” for missing information. Score the extracted facts and page attribution separately, and test a scanned copy separately if that is your actual use case.
Our Kimi PDF guide and Qwen PDF guide explain the workflows and checks. They should not be treated as a matched Kimi-versus-Qwen benchmark. Before uploading, confirm that the exact service is appropriate for the document and that you are authorized to share it.
Office deliverables: Kimi Work vs QwenWork
Kimi’s web assistant can create documents, spreadsheets and presentations. For work involving folders and desktop applications, compare Kimi Work with QwenWork. Both extend beyond a chat answer into actions and deliverables.
| Capability | Kimi Work | QwenWork |
|---|---|---|
| Product and availability | Desktop Work mode for Mac and Windows; currently described as Beta. | Web and desktop productivity platform with personal and organization routes. |
| Files and deliverables | Authorized local files, documents, spreadsheets, presentations and reports. | Local and workspace files, documents, presentations, reports, webpages and multimedia outputs. |
| External services | Skills, plugins, professional databases and WebBridge browser workflows. | Skills, connectors, browser automation and supported integrations. |
| Longer tasks | Goal Mode and Agent Swarm where included in the plan. | Multi-stage sessions and recurring scheduled tasks. |
| Desktop control | Permission-controlled local operations and supported computer-use workflows. | Computer Control is documented for macOS 14 or later with the required system permissions. |
| Billing | Eligible Kimi membership credits and feature limits. | QwenWork credits and the selected personal or enterprise plan. |
The Kimi Work overview and QwenWork documentation describe these workflows. Check the exact operating system and feature before choosing: a Windows desktop application does not establish Windows support for QwenWork’s separately documented Computer Control.
A local desktop agent is not automatically an offline model. It can access files on your computer while relying on hosted inference and external services. Desktop scheduled work can also depend on the client and computer remaining available; cloud schedules use a different execution route.
Judge the exported file. Open the DOCX, spreadsheet or presentation in the application you will actually use. Check editable text, cells and shapes, formulas, chart values, fonts and links. A slide preview cannot establish whether the exported deck is editable. After a revision, check that previously correct pages or sheets still work.
Begin an agent comparison in a copied project folder with restricted access and approval for consequential actions. Give each tool the same source files and a concrete deliverable, then inspect its changes and the final export.
Coding: Kimi Code vs Qwen Code
For repository work, compare the coding products rather than a code snippet in ordinary chat. Kimi Code is a managed coding service tied to eligible Kimi membership benefits. Qwen Code is an open-source coding agent whose model provider and inference billing are configured separately.
| Question | Kimi Code | Qwen Code |
|---|---|---|
| Where is it used? | Official CLI, Desktop and VS Code clients, plus supported external coding tools. | Terminal and documented editor or automation integrations. |
| What does it do? | Inspects and edits repository files, runs tools and assists with software tasks. | Explores codebases, edits files and runs approved commands using the configured model. |
| What pays for inference? | Eligible membership quota or supported additional usage; general Kimi API billing is separate. | The chosen provider’s plan, API billing or local infrastructure. |
| What must you record? | Membership generation and region, model access, quota and client settings. | Provider, model ID, authentication, tools and permission settings. |
For new China subscriptions, Go does not include Kimi Code; Plus and higher do. Legacy-plan rules differ, as explained in the plans section. The stable identifier kimi-for-coding is a service-facing identifier whose backend can change, not a promise of one permanent model snapshot.
Use identical clean repository snapshots, the same issue and acceptance criteria, and matching tools and permissions. Preserve the first patch, run the same tests and inspect all changed files. Compare the time needed to reach a reviewed patch, including retries and manual repair.
Our Kimi Code guide and Qwen Code guide cover the individual workflows. Installing agent software and paying for inference are separate decisions; neither a free installation nor a passing test suite guarantees unlimited usage or production-ready code.
Images, video and audio: where Qwen’s breadth matters
Qwen is the broader first ecosystem to explore when your main requirement is media creation. Its catalog includes visual understanding, image generation and editing, video generation, speech and audio routes. Select the exact consumer feature or API model instead of assuming every Qwen model supports every modality.
Kimi K3 supports image and video understanding through its API, and Kimi products can use Skills or plugins for additional media workflows. A plugin’s generation capability should not be attributed automatically to the Kimi K3 foundation model.
| Task | Useful comparison |
|---|---|
| Read a chart or photograph | Compare both supported visual routes on small labels, numbers, missing details and source attribution. |
| Analyze video | Compare temporal coverage and timestamps using the same source; document any preprocessing or tool differences. |
| Generate or edit images | Start with Qwen’s dedicated image routes and check text, object consistency, edit preservation and the exported file. |
| Generate video or use speech APIs | Explore the applicable Qwen specialist route; confirm duration, format, language support and billing units. |
qwen3.8-omni-flash accepts text, image, audio and video inputs, but its documented output is text. Its “Omni” name should not be read as a promise that the same endpoint generates every media type. See the official model page.
Kimi API vs Qwen API: models and costs
An API comparison needs exact model IDs and providers. The international Kimi API uses https://api.moonshot.ai/v1; QwenCloud’s OpenAI-compatible route uses https://maas.qwencloudapi.com/compatible-mode/v1. Credentials, balances and prices are specific to the selected platform.
For the Qwen implementation route, use our Qwen API setup guide and match the key to its provider and region.
| Model or route | Documented role | Important distinction |
|---|---|---|
kimi-k3 | Kimi flagship with a one-million-token context, visual input, tool use and structured output. | Thinking is always enabled; reasoning effort is configurable. |
kimi-k2.7-code | Coding-specialized model with a 256K context. | This API model is separate from the Kimi Code membership service. |
qwen3.8-max | QwenCloud flagship with a one-million-token context, image/video understanding and tools. | A hosted model ID is not a downloadable checkpoint name. |
qwen3.8-flash | Lower-priced QwenCloud model with a one-million-token context and visual input. | Price alone does not show whether it meets your task’s quality requirements. |
The Kimi model catalog and QwenCloud selection guide document the current routes. Use our Kimi model guide and Qwen model guide for the broader naming and deployment distinctions.
Kimi API search caveat: the K3 quickstart currently says its API web-search tooling is being updated and is not recommended for production workflows in the near term. That warning is about the developer tooling; it does not establish that consumer Kimi Search or Deep Research is unavailable.
Selected input and output rates
Verified October 13, 2026. Prices below are US dollars per one million tokens for the named first-party routes. They cover full-rate input and output only; cache operations, tools and other charges require their own billing rules.
| Model | Full-rate input / 1M tokens | Output / 1M tokens | Official pricing source |
|---|---|---|---|
kimi-k3 | $3.00 | $15.00 | Kimi API Platform |
qwen3.8-max | $2.00 | $6.00 | Qwen3.8 Max |
qwen3.8-flash | $0.15 | $0.47 | Qwen3.8 Flash |
For 500,000 full-rate input tokens and 100,000 total billable output tokens, including any reasoning, the input/output-only subtotals would be:
| Model | Input/output calculation | Illustrative subtotal |
|---|---|---|
| Kimi K3 | (0.5 × $3.00) + (0.1 × $15.00) | $3.000 |
| Qwen3.8 Max | (0.5 × $2.00) + (0.1 × $6.00) | $1.600 |
| Qwen3.8 Flash | (0.5 × $0.15) + (0.1 × $0.47) | $0.122 |
This is an input/output subtotal, not a complete request bill. Kimi K3 can incur automatic cache-write charges, with different retention tiers. QwenCloud has model-specific implicit and explicit cache rules. Cache creation or storage, search, tool execution, separate media charges, additional requests and taxes are excluded here. Apply the Kimi billing rules or QwenCloud billing rules to the actual usage.
When reasoning is already included in an API’s reported billable output total, count it once. Also compare cost per verified task: a lower token rate can be offset by longer output, failed attempts, tool calls or manual correction. These prices do not establish equal model quality.
Open-weight models and local deployment
Both ecosystems publish model weights. Kimi K3 is an exceptionally large 2.8-trillion-parameter mixture-of-experts checkpoint with native vision and a one-million-token context. Its full deployment requires substantial infrastructure; the number of parameters activated per token does not describe the storage needed for all weights.
Qwen offers more choices across model sizes and specialist families. For example, the official Qwen3.8-27B checkpoint uses Apache 2.0. That license does not apply automatically to every model carrying the Qwen name.
Choose a checkpoint that fits your hardware and read its exact license. Account for model storage, working memory, the context cache, quantization and runtime support. Checkpoint availability does not imply that a laptop can run it at a useful speed.
Self-hosting can give you more control over inference, but review the complete data path. Remote tools, model downloads, telemetry, embeddings, logs and cloud fallbacks can still involve other services. Hardware, electricity and maintenance also remain costs.
Privacy and data handling depend on the route
A paid plan, a desktop installation or an open-weight label is not a complete privacy answer. Check the service agreement that applies to your account and the files you intend to share.
| Route | What the cited documentation says | Practical implication |
|---|---|---|
| Kimi China consumer service | Its data-usage guidance permits training with securely de-identified interactions, describes an account-level opt-out and states that data is stored in mainland China. | Use the documented request process if needed; this regional guidance should not be generalized to every Kimi service. |
| International Kimi | Its separate privacy policy permits certain uses of content for model improvement and storage outside the user’s country; it does not identify one storage country. | Check the applicable legal basis and privacy rights. Do not assume the China service’s training opt-out procedure applies. |
| Kimi API | Kimi says API inputs and outputs are not used to train or improve its models. | Still review retention, uploaded-file handling, region and the applicable developer agreement. |
| Qwen consumer service | Its policy permits training and improvement using de-identified content and feedback. Its agreement allows an opt-out request at [email protected], with a feedback exception. | Feedback remains eligible for training and can retain the related conversation. An opt-out does not replace permission to share a document. |
| QwenWork international route | The international documentation says user content is stored in Singapore and describes prompts, files, outputs and usage data. | Review personal or enterprise terms, integrations and workspace permissions. |
| Desktop agents and self-hosted models | Data handling depends on local permissions, model providers, external tools and logs. | Limit accessible folders and inspect every connected service. |
Sources for these distinctions: Kimi China data usage, International Kimi Privacy Policy (updated August 13, 2026), Kimi API data handling, Qwen Privacy Policy (effective September 7, 2026), Qwen Service Agreement (effective September 10, 2026), and QwenWork privacy and security.
For confidential work, confirm authorization to share the material with the exact provider, tools and service route. Use non-sensitive sample documents while evaluating a product’s behavior.
How to choose with one useful comparison
- Choose a representative task. Use non-sensitive material and define the finished result you need.
- Match the product layer. Compare chat with chat, Work with Work, Code with Code or a named API route with another named route.
- Record the setup. Note the date, region, subscription generation, model, tools, permissions and provider.
- Define success first. Write the required facts, page references, calculations, file properties or code tests.
- Keep the first outputs. Preserve omissions, incorrect answers and broken files as well as successes.
- Verify independently. Open sources and exports, recalculate numbers and run the relevant code checks.
- Compare repair and cost. Give both the same correction request, then measure the work and expense needed to reach an acceptable result.
Use a second task when it represents a different workflow you actually need. Agreement between two assistants is not independent verification: use source documents, calculations and tests to establish correctness.
This article does not measure comparative accuracy, speed, long-document recall or coding success rates. Consumer backends, account features and model aliases can change. The recommendations identify suitable starting points from current documentation.
Frequently asked questions
Is Kimi better than Qwen?
There is no overall winner established here. Start with Kimi when an integrated research and office workflow matches your goal. Explore Qwen when media coverage, model selection or lower-priced API routes matter more. Choose using the verified result of a representative task.
Are Andante and Go the same Kimi plan?
They should not be treated as identical. Andante is a legacy China plan that existing subscribers can retain and renew; Go is a current China plan. Their Kimi Code entitlement differs: legacy Andante retains access, while Go has no included coding quota.
Does a Kimi membership include API credit?
General Kimi API Platform billing is separate from consumer membership. Eligible Kimi Code access is a membership benefit with its own service and usage rules; it does not turn the general API platform into an unlimited subscription.
Is Qwen completely free?
The ordinary consumer assistant is described as free, within its service terms and available allowances. QwenWork subscriptions, hosted APIs, provider plans and local infrastructure have separate costs. Identify the product before comparing prices.
Which is better for PDFs and coding?
Neither result is established by this documentation review. For PDFs, verify facts, missing information and page references from the same document. For repository work, compare Kimi Code with Qwen Code using the same issue and tests. A larger context window or a fluent explanation is not enough to decide.
Can I use both?
Yes, when each has a clear role and its terms fit the work. You might use Kimi for a research deliverable and a Qwen API or media model for another stage. Record the provider and product behind each output, and verify important claims independently.
Sources and update notes
The official sources are linked beside the claims, model rates and plan tables above. Key membership references checked on October 13, 2026 are:
- Kimi China membership pricing and legacy-plan rules.
- Kimi China membership benefits overview.
- Kimi international membership pricing and legacy-plan rules.
- Kimi Code eligibility, legacy plans and usage limits.
- QwenWork personal and enterprise pricing.
October 13, 2026 editorial update: replaced the outdated presentation of China’s legacy Kimi tiers with the current Go, Plus, Pro and Max menu; separated international plans; clarified legacy renewal, Code eligibility and usage limits; rechecked the remaining product, billing and data-policy references; and consolidated repeated comparison sections. This update did not include new app trials or a matched performance benchmark.