Kimi
Projects and a separate research-report workflow
by Moonshot AI
Overview
Kimi is Moonshot AI’s consumer assistant with Projects and a separate Deep Research workflow. Projects accepts PDF, DOCX, XLSX, CSV, TXT, Markdown, code and common images: up to 50 files at 100 MB each. Free lists two projects and 500 MB total project storage. Files are read on demand, not all loaded in full on every turn; Kimi Work Projects and chat Projects do not interoperate.
Deep Research plans and searches, uses code tools and produces a sourced report with a shareable HTML view and PDF/Word export. Its Researcher context is 128K including input and output, with recommended reference material plus instructions below 100K. K3’s larger API context is not Researcher’s window. The documentation does not establish that ten Project PDFs automatically transfer in full to the separate research entry.
Consumer membership names and billing evidence differ by channel. The China web pricing page, re-read on 2026-10-15, now lists Go ¥49, Plus ¥99, Pro ¥199 and Max ¥699 a month with annual totals ¥468/948/1,908/6,708 and monthly equivalents ¥39/79/159/559, and states the equivalent is not billed monthly; Andante/Moderato/Allegretto/Allegro are marked legacy plans with no published prices, so the same four monthly amounts must not be mapped back to the old names. App-store SKUs in China and the US still carry ordinary labels without a billing period, so the US Plus $19 listing cannot be promised as $19/month. Agent, Code, Work and Claw share applicable membership allowance, while developer API use is billed separately.
The consumer policy permits de-identified inputs/outputs for improvement, with requests to stop training through support, and stores personal information in China. Enterprise no-training terms must not be applied to personal accounts. We have not tested full PDF coverage, source disagreements, citation accuracy, account allowance or a complete US$20 monthly workflow.
Editor's take
A document/research candidate with specific Projects limits and documented PDF/Word research exports. These are separate workflows: file storage capacity does not establish full research coverage. China's new-plan monthly billing is now documented, but enough allowance for ten PDFs remains unconfirmed, so no complete-task or performance recommendation is made.
Our reading of who this suits, based on the sources above. Opinion, offered to help you decide - not a lab result.
Key Features
- •Projects: up to 50 files, 100 MB each; Free has two projects and 500 MB total project storage
- •On-demand file reading; chat Projects and Kimi Work Projects are separate
- •Deep Research with source-bearing reports, a shareable HTML view and PDF/Word export
- •Researcher context 128K including input/output; recommended input below 100K
- •Shared applicable consumer membership allowance across Agent, Code, Work and Claw
- •Separately metered model and search APIs, with distinct keys/endpoints from Kimi Code
Use Cases
- →Keeping a bounded set of reference files in Projects and checking which sources each answer used
- →Creating sourced research reports through the separate Deep Research entry and reviewing exported PDF/Word files
- →Developer integrations with separately budgeted Kimi model, search and extraction costs
Pros
- ✓Projects publishes concrete file-count, size and Free-storage boundaries
- ✓Deep Research documents report source links and PDF/Word export
- ✓Official documentation distinguishes consumer membership, Researcher and developer API workflows
Cons
- ✕Projects reads files on demand; ten stored PDFs are not proof that every page reaches Deep Research
- ✕China's web pricing now states plan names and billing periods; ordinary app-store SKU periods and task allowance remain unverified
- ✕Researcher’s 128K context cannot be replaced by K3’s 1M API specification
- ✕Consumer improvement/training conditions differ from enterprise terms; citation accuracy and full export quality were not tested
Frequently Asked Questions
How many files fit in Projects, and are they always fully read?
Projects permits up to 50 files at 100 MB each. Free lists two projects and 500 MB total project storage, so ten 100 MB files would exceed Free storage. Files are read on demand, not all loaded in full each turn. Kimi Work Projects and chat Projects do not interoperate.
What can Deep Research export, and how much context does it use?
The separate Deep Research entry provides sourced reports, a shareable HTML view and PDF/Word export. Researcher uses 128K context for input plus output; instructions and reference material are recommended below 100K, with typical output around 8K–16K in the documentation. Those output figures are not a guarantee or our test result, and K3’s API context does not apply.
Source: Kimi Deep Research workflow and export · Kimi Researcher context and limits
Does Plus cost US$19 per month?
The US billing period was not verified. The US App Store lists Plus $19 and Plus Annual Pass $180.99, Pro $39 and Max $99, but ordinary SKU labels lack a period. China's web pricing page, re-read on 2026-10-15, states the periods directly: Go ¥49, Plus ¥99, Pro ¥199 and Max ¥699 a month, with annual totals ¥468/948/1,908/6,708 and monthly equivalents ¥39/79/159/559 that the page says are not billed monthly. Andante/Moderato/Allegretto/Allegro are now marked legacy plans, so do not map them onto the new names or turn an annual total into a monthly payment.
Source: Kimi US iOS purchase directory · Kimi China membership pricing and plan table
Do membership, Code and API share one bill?
Membership features share a monthly credit pool, so Code consumption affects other features. New Plus and higher plans include Code with a rolling five-hour usage window and no weekly limit; Go excludes Code. Legacy plans retain both rolling five-hour and weekly limits. These are usage windows, not a five-hour weekly runtime allowance. Credits refresh monthly with no rollover. Developer API usage is billed separately.
Source: Kimi China membership pricing and plan table · Kimi API pricing page
What developer charges were verified?
CNY per million tokens: kimi-k3 cache-miss input ¥20, hit ¥2 and output ¥100; 5-minute/1-hour cache writes ¥20/¥40, with a 1,048,576-token context. kimi-k2.6 input miss/hit/output is ¥6.5/¥1.1/¥27. File extraction/storage is temporarily free, but model input text is charged. REST search Basic/Pro is ¥0.01/¥0.015 per call and successful nonempty fetch ¥0.01. Legacy $web_search is ¥0.03/call; on 22 October 2026 the vendor page still labels it a legacy method and says it is expected to go offline on 20 October 2026, without saying it has gone offline. That legacy price does not cover every current search route.
Source: Kimi official API token rates (Markdown) · Kimi current API search rates · Kimi consumer-help API pricing caveat
Can US$20 cover ten PDFs, Chinese citations, disagreements and an exported table?
Not confirmed for the full task. The China membership page documents monthly CNY billing, but the applicable region/channel quote for this user’s US$20 budget and sufficient allowance are unestablished. Ten files fit the Projects count only subject to per-file and total storage limits. Complete reading, transfer to separate Deep Research, original-page citation accuracy and the required table output were not tested. Documented report export does not prove the full workflow.
Source: Kimi Projects file and storage limits · Kimi web membership pricing — verification incomplete · Kimi Deep Research workflow and export
Are personal uploads excluded from model training?
The consumer policy permits de-identified input/output for model improvement; users can ask support to stop training use. Personal information is stored in China. Enterprise default no-training terms are not a promise for free or personal memberships.
Source: Kimi consumer privacy policy
Task guides for this product
Key facts and where they come from
- PriceDocumented
- The China membership page lists Go CNY 49, Plus 99, Pro 199 and Max 699 per month. Annual totals are CNY 468/948/1,908/6,708; the monthly equivalents 39/79/159/559 are comparison figures, not monthly payments. Andante, Moderato, Allegretto and Allegro are legacy plans for existing subscribers, not the new plan names. The page labels its pricing illustration as captured on October 12, 2026. Historical registered scope from 2026-09-23; the entire statement was not re-read on 2026-10-15: Plus $19, Annual Pass $180.99, Pro $39 and Max $99 alongside older names. Ordinary SKU periods and web equivalence were not verified; $19 must not be labelled monthly from this listing alone.Source: Kimi China membership pricing and plan table · Kimi US iOS purchase directory
- AvailabilityNot established
- The membership help explicitly scopes its current table to China and directs eligibility to the purchase page. It does not establish where all consumer plans are distributed or actually reachable, so those geographic access conditions remain unknown.Source: Kimi China membership pricing and plan table
- Commercial useNot established
- The service agreement lists personal learning/research/appreciation/life/entertainment and allocates output rights subject to law, third-party rights and compliance. It does not establish permission to use the actual paid-module output commercially.Source: Kimi user agreement — model use (v2)
- Training on your dataDocumented
- Kimi states that de-identified inputs and outputs may be used to improve its models, that a user can ask support to stop that training use, and that personal information is stored in China; enterprise no-training terms do not transfer to personal plans.Source: Kimi consumer privacy policy
- API billingDocumented
- Consumer membership and the developer API are separate meters: Agent, Code, Work and Claw share the applicable membership allowance, while the platform prices model tokens (kimi-k3 miss/hit/output at CNY 20/2/100 per million with cache writes at 20/40, kimi-k2.6 at 6.5/1.1/27) and REST search per call (CNY 0.01/0.015, with a successful non-empty fetch at 0.01); the legacy $web_search rate of CNY 0.03 per call remains listed as a legacy method; the page re-read on 22 October 2026 still says it is expected to go offline on 20 October 2026, not that it has, and this rate does not cover every current search route.Source: Kimi official API token rates (Markdown) · Kimi current API search rates · Kimi shared membership allowance
- Export formatsDocumented
- Separate research entry uses search/planning/code tools and provides sourced reports, HTML sharing and PDF/Word export. Historical scope read on 2026-10-12, not rechecked in full on 2026-10-15: Complete transfer of ten Project files into this separate workflow was not established by the reviewed materials.Source: Kimi Deep Research workflow and export · Kimi Researcher context and limits
- Usage limitsDocumented
- Projects permits up to 50 files at 100 MB each; Free has two projects and 500 MB total storage. Files are read on demand, and chat Projects and Kimi Work Projects do not interoperate. Researcher has 128K input-plus-output context, with recommended input below 100K; this is not a consumer-wide or API context entitlement. Historical scope read on 2026-10-12, not rechecked in full on 2026-10-15: Researcher context is 128K input plus output; recommended input below 100K, typical output around 8K–16K. K3 context must not be transferred to Researcher. Complete transfer of ten Project files into this separate workflow was not established by the reviewed materials. Membership features share a monthly credit pool; using credits on one feature affects the others. Monthly and annual subscriptions refresh credits monthly with no rollover. Go excludes Kimi Code; new Plus and higher plans include it with a rolling five-hour usage window and no weekly limit. Legacy plans retain both rolling five-hour and weekly limits. These are usage windows, not five hours of use per week, and exhausting the shared monthly pool also affects Code.Source: Kimi Projects file and storage limits · Kimi Researcher context and limits · Kimi China membership pricing and plan table
Official sources
The vendor pages we read for the claims on this page, what each one states, and the day we read it.
- Kimi Projects file and storage limits ↗
Historical registered scope from 2026-10-12; the entire statement was not re-read on 2026-10-15: Up to 50 files, 100 MB each; Free two projects and 500 MB total project storage. Files are read on demand. Chat Projects and Kimi Work Projects do not interoperate. Official body read on 2026-10-15, limited to this scope: Projects permits 50 files of 100MB each; Free has 2 projects/500MB storage. Files are read on demand, not fully preloaded each turn; chat Projects and Kimi Work Projects are separate. Projects permits up to 50 files at 100 MB each; Free has two projects and 500 MB total storage. Files are read on demand, and chat Projects and Kimi Work Projects do not interoperate.
Read on October 15, 2026
- Kimi Deep Research workflow and export ↗
Separate research entry uses search/planning/code tools and provides sourced reports, HTML sharing and PDF/Word export. It uses membership allowance; the 5–10% per-task estimate applies to old Moderato, not all plans. Vendor output/speed estimates are not this site’s tests.
Read on September 23, 2026
- Kimi Researcher context and limits ↗
Historical registered scope from 2026-10-12; the entire statement was not re-read on 2026-10-15: Researcher context is 128K input plus output; recommended input below 100K, typical output around 8K–16K. K3 context must not be transferred to Researcher. Complete transfer of ten Project files into this separate workflow was not established by the reviewed materials. Official body read on 2026-10-15, limited to this scope: Researcher context is128K input+output, roughly60, 000–100, 000 Chinese characters; recommended input below100K. This is Researcher, not a consumer-wide or API context promise. Historical scope read on 2026-10-12, not rechecked in full on 2026-10-15: Complete transfer of ten Project files into this separate workflow was not established by the reviewed materials. Researcher has 128K input-plus-output context, with recommended input below 100K; this is not a consumer-wide or API context entitlement. Historical scope read on 2026-10-12, not rechecked in full on 2026-10-15: Researcher context is 128K input plus output; recommended input below 100K, typical output around 8K–16K. K3 context must not be transferred to Researcher. Complete transfer of ten Project files into this separate workflow was not established by the reviewed materials.
Read on October 15, 2026
- Kimi China membership pricing and plan table ↗
This help explicitly scopes the new plan table to China and distinguishes monthly billing from upfront annual totals. It directs purchase eligibility to checkout; it is not a distribution or reachability list for every consumer plan. Historical registered scope read on 2026-10-15, not rechecked in full on 2026-10-15: The China membership page lists Go CNY 49, Plus 99, Pro 199 and Max 699 per month. Annual totals are CNY 468/948/1,908/6,708; the monthly equivalents 39/79/159/559 are comparison figures, not monthly payments. Andante, Moderato, Allegretto and Allegro are legacy plans for existing subscribers, not the new plan names. The page labels its pricing illustration as captured on October 12, 2026. Membership features share a monthly credit pool; using credits on one feature affects the others. Monthly and annual subscriptions refresh credits monthly with no rollover. Go excludes Kimi Code; new Plus and higher plans include it with a rolling five-hour usage window and no weekly limit. Legacy plans retain both rolling five-hour and weekly limits. These are usage windows, not five hours of use per week, and exhausting the shared monthly pool also affects Code.
Read on October 15, 2026
- Kimi shared membership allowance ↗
Agent, Code, Work and Claw share applicable monthly allowance, with token-related consumption; the old table scopes million-token context to Allegro.
Read on September 23, 2026
- Kimi China iOS purchase directory ↗
Historical registered scope from 2026-09-23; the entire statement was not re-read on 2026-10-15: Go ¥49, annual ¥468; Plus ¥99, annual ¥948; Pro ¥199 alongside older names. Ordinary SKU labels do not establish billing periods; annual passes are annual totals. Official body read on 2026-10-15, limited to this scope: The Chinese regional App Store returns a Kimi listing. This is app distribution evidence, not proof of account registration or web/service reachability in a region.
Read on October 15, 2026
- Kimi US iOS purchase directory ↗
Historical registered scope from 2026-09-23; the entire statement was not re-read on 2026-10-15: Plus $19, Annual Pass $180.99, Pro $39 and Max $99 alongside older names. Ordinary SKU periods and web equivalence were not verified; $19 must not be labelled monthly from this listing alone. Official body read on 2026-10-15, limited to this scope: The US regional App Store returns Kimi and lists English plus15 more languages. Store distribution and 16 interface languages do not establish service reachability or a monthly billing period.
Read on October 15, 2026
- Kimi web membership pricing — verification incomplete ↗
Dynamic plan text was not readable in this review; new-user checkout, allowances and annual pricing remain unverified, rather than being absent or undisclosed by the vendor.
Read on September 23, 2026
- Kimi Code plan and limit errors ↗
Distinguishes legacy Andante/Moderato/Allegretto/Allegro from Go/Plus/Pro/Max/Ultra; old plans have weekly limits, new ones do not have that weekly limit. Code also has five-hour and shared monthly allowance boundaries.
Read on September 23, 2026
- Kimi official API token rates (Markdown) ↗
CNY/million: K3 miss/hit/output 20/2/100 and 5-min/1-hour cache writes 20/40, context 1,048,576; K2.6 6.5/1.1/27. File extraction/storage is temporarily free but extracted model input is billed.
Read on September 23, 2026
- Kimi API pricing page ↗
The web parser did not expand its custom table; the same official Markdown source was readable and supplied the token rates. Consumer membership is separate.
Read on September 23, 2026
- Kimi current API search rates ↗
REST Basic/Pro search CNY 0.01/0.015 per call; successful nonempty fetch 0.01. Legacy $web_search is 0.03/call. Re-read on 22 October 2026, the page still lists $web_search under "legacy method" (存量方式) and says it is "expected to go offline on October 20, 2026" (预计), without saying it has gone offline; retirement is not confirmed and this site does not treat the tool as retired.
Read on October 22, 2026
- Kimi consumer-help API pricing caveat ↗
Its blanket CNY 0.03/search statement matches the legacy route and does not represent every current REST search mode; use current platform pricing per endpoint.
Read on September 23, 2026
- Kimi consumer privacy policy ↗
De-identified inputs/outputs may improve models; users may contact support to stop training use. Personal information is stored in China; enterprise no-training terms must not be transferred to personal plans.
Read on September 23, 2026
- Kimi user agreement — model use (v2) ↗
Section 1.3 limits listed service purposes to personal learning/research/appreciation/daily life/entertainment; section 5.2 assigns content rights subject to law, third-party rights and agreement compliance. This does not establish commercial permission for the actual output/paid module. Historical registered scope read on 2026-10-15, not rechecked in full on 2026-10-15: The agreement effective 31 August 2026 lists personal learning, research, appreciation, daily life and entertainment in section 1.3. Section 5.2 allocates input/output rights to the user subject to law, third-party rights and compliance; ownership alone does not establish a commercial-output licence. Section 5.3 permits model optimisation using inputs, outputs and feedback and provides a contact opt-out. Specific-module and paid-service agreements prevail in a conflict. Whether the actual output and paid module have the required commercial-use permission remains unestablished.
Read on October 15, 2026
Saved in this browser only; not synced across devices.
Public signals
Read on 2026-09-20 from the sources linked below.
- Wikipedia interest
- 24031 views per monthArticle: Kimi (AI) (2026-07..2026-08)View source
Basic Info
- Pricing
- Freemium
- Category
- Chat Assistants
- Launch Date
- November 16, 2023
- Maker
- Moonshot AI
- Last verified
- September 23, 2026
- Content updated
- October 22, 2026
Compare with other tools in Chat Assistants
| Product | Maker | Pricing | Launched | Shared tags |
|---|---|---|---|---|
| Doubao | ByteDance | Freemium | 2023-08 | Chat, Research |
| Qwen Studio | Alibaba | Free | 2023-04 | Chat, Research |
| Claude | Anthropic | Freemium | 2023-03 | Chat, Long context |
| Mistral Vibe | Mistral AI | Freemium | 2024-02-26 | Chat |
| Grok | xAI | Freemium | 2023-11-03 | Chat |
| DeepSeek | DeepSeek | Free | 2023-11-02 | Chat |



