SKYW
Back to all projects

PRE-IPO / PUBLIC PROJECT PROFILE

Moonshot AIKimi

From model intelligence to real-world work.

A closer look at Moonshot AI and Kimi: foundation model research, agentic applications and a developer ecosystem, viewed through a long-term investment lens.

This profile presents a project and investment perspective. It does not confirm fund ownership, an investment allocation or an official partnership with Moonshot AI.

K3 total parameters
2.8T
Context window · tokens
1M
Mixture-of-experts architecture
MoE

Specifications follow official public materials, not product revenue or investment returns. [1][3]

Where research meets products and applications

Moonshot AI is the company behind Kimi. Its public offering connects model research with research, documents, spreadsheets, presentations and coding. Our focus is not a single leaderboard result, but whether capabilities become useful in sustained workflows. [1]

01 / Model Evolution

From tool use to multimodal, long-horizon work

Follow the public model family to understand research direction and its connection to products. This is a capability sequence, not a financial growth chart.

  1. K2

    An open agentic foundation

    Tool use is central to the official introduction, with separate Base and Instruct models. [7]

  2. K2.5

    Vision and parallel collaboration

    Native multimodality, visual coding and Agent Swarm feature in the public release. [8]

  3. K2.6

    Long-horizon engineering

    Public materials extend the focus on coding, long-horizon execution and agent coordination. [9]

  4. K3

    Connecting models and products

    K3 continues the research trajectory through several product interfaces. [2]

02 / Product Ecosystem

One model foundation. Multiple ways to work.

A capability map based on official product descriptions, not a revenue breakdown. [2]

Kimi K3FOUNDATION MODEL

Kimi

An agentic workspace for users

Kimi Work

A desktop agent for knowledge work

Kimi Code

Coding tools for terminal and IDE

Kimi API

Model access for developers

03 / Technology

Three building blocks of capability

Explore publicly documented work on long context, computation and multimodal collaboration.

KDA + AttnRes

Context and information flow

Architectural work on information flow across sequence length and depth. [3]

Stable LatentMoE

Sparse computation

A mixture-of-experts approach to large-scale model computation. [3]

Vision + Agents

Multimodality and tools

Research connecting vision, reasoning and long-horizon tasks. [4]

Official evaluations also disclose limitations. Results depend on tasks, tooling and usage; parameter scale or a benchmark cannot be converted directly into investment value. [2]

From research releases to an open ecosystem

  1. 2026.07

    K3 introduction

    Official model positioning, technical directions and product access. [2]

  2. 2026.07.27

    Open model and research release

    Weights, a technical report and infrastructure released, subject to applicable licenses. [3]

04 / Reading the Technology

What do these capabilities mean in practice?

An interpretation of technical concepts, not a guarantee of results on every task.

Context: the material available to a task

Long context creates room for documents, code and task history. Capacity is not the same as comprehension: retrieval, omissions, source accuracy and completion quality still need testing.

Agents: from answers to actions

Agents plan, use tools and adapt to results. Evaluation must cover the process as well as the answer: traceability, permission boundaries and safe stopping when something goes wrong.

Open weights: more choice, continuing responsibility

Open weights offer research and deployment choices, not cost-free operations or automatic compliance. Licensing, compute, operations, security and model upgrades remain part of the decision.

05 / Applications

Putting capabilities into working contexts

Four application perspectives informed by Kimi Work and Kimi Code. These are not claims of SKYW or partner deployments, nor quantified productivity promises. [5][6]

RESEARCH

Research and information organization

Kimi Work describes local-file handling and browser research. [5] Investment teams could evaluate public-material organization first, without treating model output as an investment conclusion.

Output to evaluate
Sourced summaries and open questions
Human review focus
Dates, provenance, definitions and evidential support
WORKSPACE

Document and data deliverables

The product describes exporting work to presentations, spreadsheets and documents. [5] An enterprise evaluation should examine editability, reusable structure and the effort needed for corrections.

Output to evaluate
Briefings, structured tables and presentation drafts
Human review focus
Formulas, units, completeness and file versions
ENGINEERING

Software engineering and code collaboration

Kimi Code describes writing, debugging, refactoring and codebase analysis. [6] A useful evaluation asks whether it follows existing engineering conventions, rather than measuring generated code volume.

Output to evaluate
Prototypes, proposed fixes and test drafts
Human review focus
Tests, security, dependency licenses and maintenance
OPERATIONS

Recurring tasks and operational workflows

Kimi Work describes scheduled and background tasks. [5] For recurring work, evaluation should move beyond a successful run to monitoring, records and human escalation when exceptions arise.

Output to evaluate
Recurring briefings, information digests and update checks
Human review focus
Permissions, alerts, logs and human takeover

A research task, from brief to reviewable output

  1. 01

    Set boundaries

    Define the task, sources, permissions and acceptance criteria.

  2. 02

    Organize and execute

    Break down questions, organize material and keep records.

  3. 03

    Verify with people

    Review citations, figures, reasoning and exceptions.

  4. 04

    Deliver and improve

    Deliver traceable work and record corrections and feedback.

An illustrative evaluation workflow, not a Kimi interface, a customer deployment or an automated investing system.

06 / Investment Perspective

An investment lens: how capability becomes value

A research framework, not fund terms, holdings disclosure or a performance forecast.

01

Research durability

Assess iteration, talent and engineering execution beyond a single model launch.

02

Commercial quality

Examine paying demand, retention, delivery and inference costs, supported by financial and operating evidence.

03

Structure and risk discipline

Separate business quality from deal terms, including equity rights, valuation evidence, concentration and liquidity.

07 / Commercial Research

Commercialization: from usage to revenue quality

A research framework across end-user products, developer access and organizational adoption, not reported financial segments. No private revenue, ARR, customer counts or valuations are disclosed.

End-user products: reasons to keep paying

For a workspace such as Kimi, visits and downloads are only a starting point. Research should examine repeat valuable tasks, conversion, renewal and service costs, separating initial interest from durable demand.

Developer access: usage that retains customers

API research should cover price, reliability, latency, tool compatibility and switching costs. More calls do not automatically improve profit: cost per completed task, concentration and sustained usage also matter.

Organizations: moving beyond a pilot

Organizational adoption involves business, technology and risk teams. Integration, data permissions, maintenance and support help distinguish a repeatable everyday tool from a compelling demonstration.

Questions that take the investment research further

01

Can an advantage endure?

Evaluate research cadence, engineering and customer experience together, not one leaderboard result.

02

What investment sustains growth?

Separate training, inference, acquisition and operations, and assess funding needs against verified accounts.

03

Are the equity rights clear?

Assess the business separately from rights, dilution, transfer restrictions and access to information.

04

Is the exit assumption prudent?

Consider different holding periods and market conditions; a Pre-IPO label is not a filing or listing guarantee.

08 / Information Access

Public context and controlled investment materials

Specific investment materials require authorization, suitability and applicable jurisdictional checks. Partner status does not itself establish investor eligibility. No subscription is offered on this page.

Public website

  • Project identity and official product information
  • AI and private-technology investment themes
  • Research framework, key risks and contact details

Subject to separate disclosure clearance

  • Fund / manager identity and partnership or distribution roles
  • Investment structure, fees and subscription documents
  • Deal valuation, allocations, liquidity and exit arrangements
09 / Key Risks

Understand the opportunity and the risks

Technology & competition

Model evolution, compute supply and competition can affect product and commercial outcomes.

Private-market liquidity

Private equity may lack trading or exit opportunities for extended periods and may lose all invested capital.

Valuation & concentration

Valuations are uncertain; single-project concentration can amplify risk.

Regulation & listing uncertainty

Data, AI and cross-border investment rules may change. Listing timing, approval and exit are not guaranteed.

10 / Investor & Partner Questions

Questions from investors and partners

Clarifying what this profile covers and what requires a separate discussion.

01How do Moonshot AI, Kimi and K3 relate?

Moonshot AI is the company, Kimi its product brand and K3 part of the public model family. Keep company, product and model analysis distinct: specifications are not financial results, and product experience does not directly establish equity value. [1][2]

02Why are there no valuation-discount or revenue charts?

Such information needs disclosure authority, definitions, dates and verifiable evidence. Financing valuations, equity terms and revenue definitions may not be comparable. This page prioritizes public business information rather than inferring investment safety from model rankings.

03Does this profile confirm an existing fund investment?

No. A project profile does not establish a holding, available allocation, partnership or distribution arrangement. Any specific investment relationship must be supported by valid formal documents provided with authorization.

04Where can a partner discussion begin?

Discussions can cover AI use cases, research, industry resources and diligence questions. Fund documents or deal arrangements require separate identity, authorization and eligibility checks. A general enquiry does not start a subscription.

Speak with our investment team

Speak with our investment team