# Kimi (chatbot)

> Source: https://aiwiki.ai/wiki/kimi
> Updated: 2026-11-01
> Fact-checked: 2026-10-30
> Categories: Chinese AI, Conversational AI, Large Language Models
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Kimi (chatbot)." aiwiki.ai, 1 Nov 2026. https://aiwiki.ai/wiki/kimi
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**Kimi** is an [artificial intelligence](https://aiwiki.ai/wiki/artificial_intelligence) assistant and [chatbot](https://aiwiki.ai/wiki/chatbot) developed by [Moonshot AI](https://aiwiki.ai/wiki/moonshot_ai). Moonshot announced the service in China in October 2023 with support for prompts of about 200,000 Chinese characters, and opened it to general public access in November 2023.[1][2] The product later expanded from text chat and document analysis into web search, multimodal input, coding tools, research workflows, and agents.[5][6]

<figure class="w-56">
<img src="/images/site-2026-10-28-moonshot_ai-logo-da478502116f-tile.png" alt="Moonshot AI logo" width="480" height="134" />
<figcaption>Moonshot AI logo, from the official website <a href="https://moonshot.ai/">moonshot.ai</a>. Trademark of its owner.</figcaption>
</figure>

The name Kimi also refers to the model family that powers the hosted assistant. That family includes [Kimi K1.5](https://aiwiki.ai/wiki/kimi_k1_5), [Kimi K2](https://aiwiki.ai/wiki/kimi_k2), [Kimi K2 Thinking](https://aiwiki.ai/wiki/kimi_k2_thinking), [Kimi K2.5](https://aiwiki.ai/wiki/kimi_k2_5), [Kimi K2.6](https://aiwiki.ai/wiki/kimi_k2_6), [Kimi K2.7 Code](https://aiwiki.ai/wiki/kimi_k2_7_code), [Kimi K3](https://aiwiki.ai/wiki/kimi_k3), and [Kimi K3.1 Pro](https://aiwiki.ai/wiki/kimi_k3_1_pro). The models differ in architecture, modality, context length, deployment format, and license. A capability advertised for one model or interface therefore does not necessarily apply to every version of Kimi.

Kimi K3 became the flagship model in the hosted assistant in July 2026. On October 23, 2026, Moonshot released Kimi K3.1 Pro, a post-trained version of K3; as of October 30, 2026, Moonshot's API documentation describes K3.1 Pro as its most capable flagship and K3 as the previous-generation flagship.[28][30] Moonshot released K3 through its consumer and developer products on July 16 and published its full weights on July 27.[17][19] The K3 [model card](https://aiwiki.ai/wiki/model_card) describes a sparse [mixture-of-experts](https://aiwiki.ai/wiki/mixture_of_experts) model with 2.8 trillion total parameters, 104 billion activated parameters per token, native image input, and a [context window](https://aiwiki.ai/wiki/context_window) of 1,048,576 tokens.[17] These are architectural specifications published by the developer, not independent measures of output quality.

## History

### Launch and long-context development

Moonshot introduced Kimi Chat in October 2023 alongside its first [large language model](https://aiwiki.ai/wiki/large_language_model). Contemporary coverage from the state-run *Beijing Daily* described support for about 200,000 Chinese characters in one prompt.[1] The service moved from invitation-based testing to general public access in November.[2] Character counts and token counts are not interchangeable: [tokenization](https://aiwiki.ai/wiki/tokenization) depends on language and content, and later product documentation reports limits in tokens rather than Chinese characters.

In March 2024, Moonshot announced an invited beta that increased the maximum input to about two million Chinese characters. The rollout was initially limited rather than a universal change to every Kimi account.[3] The company subsequently described the infrastructure behind long-context serving in the Mooncake systems paper. Mooncake separates prompt processing from token generation and treats the key-value cache as a distributed resource that can be reused across requests. In production traces from Kimi, the authors reported serving 75 percent more requests under their service-level objectives than a comparison architecture.[4] That result was reported by Moonshot researchers on their own workload and should not be read as an independent comparison of chatbot quality.

Long context became a recurring feature of the model family, but the advertised limit has varied by model and access method. As of October 30, 2026, Kimi's capability page lists about 128,000 tokens per turn for K2.6 in the consumer chat interface and a one-million-token context for K3 with the highest membership tier, while the API documentation lists a 256,000-token context for its remaining K2-series endpoints. K3 is documented with a one-million-token context in both its model card and API materials.[7][17][30][43] Users therefore need to consult the documentation for the specific model and interface they are using.

### From chat to agents

Kimi initially centered on chat, search, and document reading. Its current help documentation describes web and mobile access, built-in web search, deep reasoning modes, file analysis, and an API. Supported uploads include common document, spreadsheet, presentation, image, text, and video formats, subject to per-file and account limits.[5][6] Moonshot also offers specialized interfaces including Kimi Code and agent products that can use tools to create documents, analyze data, or work across multiple steps.[19] By 2026 its help center distinguished four products: the Kimi web and mobile assistant, the [Kimi Work](https://aiwiki.ai/wiki/kimi_work) desktop agent, the Kimi Code coding agent, and Kimi Claw, a cloud automation service for always-on bots.[44]

Moonshot's Deep Research feature uses a separate research-agent workflow to search, evaluate, and synthesize sources. The company reports operational averages for the number of searches and pages inspected, but those figures describe its own service telemetry rather than a guaranteed workload for every query.[20] The product should also not be confused with a conventional [search engine](https://aiwiki.ai/wiki/search_engine): generated reports can omit sources, misread documents, or make unsupported inferences, so important claims still require checking against the cited material.

### 2026 product timeline

The table lists dated 2026 changes to Kimi's models, apps, and API that Moonshot documented in its own changelogs, help pages, and announcements. It is selective rather than a complete release log.

| Date (2026) | Product | Change |
| --- | --- | --- |
| January | Models | [Kimi K2.5](https://aiwiki.ai/wiki/kimi_k2_5) introduced as an open-weight multimodal agentic model.[13][14] |
| April | Models | [Kimi K2.6](https://aiwiki.ai/wiki/kimi_k2_6) released, with weights on Hugging Face.[15][16] |
| May | Kimi Code | Kimi Code CLI v0.1.0 released, a rewrite of the earlier Python kimi-cli in TypeScript and Node.js.[31] |
| May 25 | API | The original `kimi-k2` series API endpoints were discontinued.[30] |
| June 3 | Kimi Work | Kimi Work launched in beta as the desktop agent mode of the Kimi client.[33] |
| June 12 | Models | Kimi K2.7 Code released and open-sourced as a coding model.[31][32] |
| July 9 | Kimi Code | A HighSpeed tier and Extra Usage, a balance subscribers can draw on after their membership quota runs out, became available in Kimi Code.[31] |
| July 16 | Models | Kimi K3 made available in Kimi, Kimi Code, and the API; full weights followed on July 27.[17][19] |
| August 31 | API | The `kimi-k2.5` and `moonshot-v1` API models were discontinued.[30] |
| September 4 | Kimi Work | Version 3.2.5 added Remote Control, letting a phone control Kimi Work on a desktop computer.[34] |
| September 11 | Models | Kimi K2.8 Preview rolled out in Kimi Code under the existing `kimi-for-coding` model ID.[31] |
| September 17 | Kimi Code | Kimi Code Desktop released for macOS and Windows, alongside Kimi Code CLI 2.0.[31] |
| October 14 | Kimi Code | Kimi Code Desktop added Linux packages and Remote Control.[31] |
| October 23 | Models and apps | Kimi K3.1 Pro and K3.1 Pro Highspeed launched in Kimi, Kimi Work, Kimi Code, and the API; 3D Make went live on kimi.ai and the Kimi app.[28][29] |
| October 29 | Kimi Work | Version 3.2.22 added an "Ask user" step that confirms key information before a task runs, and Extra Usage packs.[34] |
| October 30 | 3D Make | Moonshot announced 3D Make for kimi.ai and Kimi Work.[35] |

Moonshot's API model list also says that, from October 31, 2026, K2-series models will no longer be available to newly registered API users, and that it plans to retire all K2-series models by the end of December 2026.[30]

<figure class="embedvideo">
<iframe src="https://www.youtube-nocookie.com/embed/VtzELsWFXV4" title="Introducing 3D Make in Kimi"></iframe>
<figcaption>"Introducing 3D Make in Kimi", official video from Kimi AI, published October 30, 2026. <a href="https://www.youtube.com/watch?v=VtzELsWFXV4">Watch on YouTube</a>.</figcaption>
</figure>

## Model family

The table summarizes the major public model releases. Dates refer to the public release or introduction described by Moonshot, not necessarily the date of a later technical paper.

| Model | Introduction | Documented characteristics |
| --- | --- | --- |
| [Kimi k1.5](https://aiwiki.ai/wiki/kimi_k1_5) | January 2025 | Multimodal reasoning model trained with [reinforcement learning](https://aiwiki.ai/wiki/reinforcement_learning); the technical report describes reinforcement learning with contexts up to 128,000 tokens.[8] |
| [Kimi K2](https://aiwiki.ai/wiki/kimi_k2) | July 2025 | Text-focused sparse model with 1 trillion total parameters and 32 billion activated parameters; the report describes pretraining on 15.5 trillion tokens.[9][10] |
| Kimi K2 Instruct 0905 | September 2025 | Updated K2 instruction checkpoint with a documented increase from 128,000 to 256,000 tokens and changes aimed at coding and agent use.[11] |
| [Kimi K2 Thinking](https://aiwiki.ai/wiki/kimi_k2_thinking) | November 2025 | Reasoning and tool-use checkpoint based on K2 with a 256,000-token context in the published [model card](https://aiwiki.ai/wiki/model_card).[12] |
| [Kimi K2.5](https://aiwiki.ai/wiki/kimi_k2_5) | January 2026 | Open-weight multimodal agentic model that combines text and vision and introduces Moonshot's Agent Swarm approach.[13][14][19] |
| [Kimi K2.6](https://aiwiki.ai/wiki/kimi_k2_6) | April 2026 | [Multimodal model](https://aiwiki.ai/wiki/multimodal_model) focused on long-horizon coding and tool use; Moonshot documented a 262,144-token evaluation context.[15][16][19] |
| [Kimi K2.7 Code](https://aiwiki.ai/wiki/kimi_k2_7_code) | June 2026 | Coding model with the K2 shape of 1 trillion total and 32 billion activated parameters and a 256,000-token context, released under a modified MIT license.[32] |
| [Kimi K3](https://aiwiki.ai/wiki/kimi_k3) | July 2026 | Native multimodal sparse model with 2.8 trillion total parameters, 104 billion activated parameters, and a 1,048,576-token context.[17] |
| Kimi K2.8 Preview | September 2026 | Model rolled out in Kimi Code with a context of up to one million tokens on all membership tiers; Moonshot describes its performance as close to K3.[31] |
| [Kimi K3.1 Pro](https://aiwiki.ai/wiki/kimi_k3_1_pro) | October 2026 | Post-trained version of K3 with substantially larger post-training compute, released with a faster Highspeed variant.[28] |

### Kimi k1.5

Moonshot introduced Kimi k1.5 in January 2025 as a multimodal reasoning model. Its technical report describes a training pipeline built around long-context [reinforcement learning](https://aiwiki.ai/wiki/reinforcement_learning), partial rollouts, and methods for distilling long reasoning traces into shorter responses. The paper reports results on mathematics, coding, and vision benchmarks and compares the model with then-current proprietary systems.[8] Those numbers are author-reported experiments. They are useful for understanding the training goals, but differences in prompts, sampling, tools, and evaluation harnesses limit direct comparisons across organizations.

### Kimi K2 and K2 Thinking

Kimi K2 moved the family to a much larger sparse architecture. The published report specifies 1 trillion total parameters, 32 billion active parameters, and pretraining on 15.5 trillion tokens. It also introduces MuonClip, a modification intended to stabilize training at that scale.[9] Moonshot released base and instruction checkpoints, code, and configuration files through its public repository.[10]

The September 2025 K2 Instruct 0905 checkpoint extended the documented context window to 256,000 tokens and targeted coding and agent tasks.[11] K2 Thinking, released in November, added a reasoning mode intended to maintain planning across long sequences of tool calls. Moonshot published benchmark results and demonstrations of 200 to 300 sequential tool calls.[12] Those demonstrations establish the intended operating mode, not a guarantee that the model will complete arbitrary long workflows correctly.

### Kimi K2.5 and K2.6

Kimi K2.5 combined the K2 language backbone with native visual input and agent-oriented post-training. Its technical report describes joint optimization of text and vision tasks and an Agent Swarm system that decomposes work among parallel subagents.[13] The checkpoint, inference guidance, and evaluation configuration were released publicly.[14] Agent Swarm is an orchestration layer around model calls; it does not mean that the underlying [neural network](https://aiwiki.ai/wiki/neural_network) contains multiple independently trained agents.

Kimi K2.6 followed in April 2026. Moonshot presented it as an improvement in long-horizon coding, tool use, and multimodal work, and released weights through [Hugging Face](https://aiwiki.ai/wiki/hugging_face).[15][16] The company said its Agent Swarm configuration could coordinate as many as 300 subagents and 4,000 steps, compared with smaller limits in the K2.5 configuration.[15] These are published system limits. They do not independently establish reliability, speed, or cost for a particular task.

### Kimi K2.7 Code and K2.8 Preview

Moonshot released Kimi K2.7 Code on June 12, 2026 as a coding model and published its weights. The model card gives the K2 shape of 1 trillion total and 32 billion activated parameters with a 256,000-token context, under a modified MIT license.[32] Moonshot's changelog reports gains over K2.6 on several coding and agent benchmarks and about 30 percent lower reasoning-token use; these are the developer's own measurements.[31]

On September 11, 2026, Moonshot rolled out Kimi K2.8 Preview in Kimi Code. The model kept the `kimi-for-coding` ID, offered low, high, and max thinking levels, and provided a context of up to one million tokens on all membership tiers. Moonshot said its coding and agent performance was close to K3, with more efficient thinking than K2.7 Code.[31] Moonshot's changelog does not give a parameter count or a weights release for K2.8 Preview.

### Kimi K3

Moonshot made K3 available in the hosted assistant, Kimi Code, and its API on July 16, 2026, then released the model weights on July 27.[17][19] Its architecture substantially increases both the total and active parameter counts relative to K2. The model card lists 896 experts, 16 selected experts per token, two shared experts, Kimi Delta Attention and gated [multi-head latent attention](https://aiwiki.ai/wiki/multi-head_latent_attention) layers, a MoonViT-V2 visual encoder, and quantization-aware training with MXFP4 weights and MXFP8 activations.[17]

K3's public benchmark table combines developer-run evaluations, third-party leaderboard values, different agent harnesses, and several tool-assisted settings. Moonshot discloses many of these conditions in its model card.[17] The scores should therefore be attributed to their individual setup rather than summarized as proof that K3 is categorically better than another model.

Demand after the hosted launch temporarily exceeded Moonshot's planned service capacity. The Associated Press reported that the company paused new subscriptions while continuing to serve existing subscribers.[21] This was a capacity-management decision during rollout, not evidence about the model's accuracy.

On October 23, 2026, Moonshot made Kimi K3.1 Pro, a post-trained version of K3, and a faster K3.1 Pro Highspeed variant available on Kimi.ai and the Kimi mobile apps, Kimi Work, Kimi Code, and the Kimi API. Its launch announcement said open weights would follow by November 4, 2026.[28][29]

## Access and use

Kimi is available as a hosted consumer service and through developer interfaces. The consumer product supports conversations, web search, file analysis, and agent workflows. The API uses request formats compatible with widely used chat-completions conventions and exposes different Kimi model endpoints.[5][24] Compatibility at the request level does not imply that every parameter, tool, or output field behaves identically across providers.

Model selection affects context limits and modality. As of October 30, 2026, the API model list offers K3.1 Pro and K3 with one-million-token contexts, a K3.1 Pro Highspeed variant, and K2.7 Code and K2.6 with 256,000-token contexts, plus a K2.7 Code Highspeed variant. The original `kimi-k2` endpoints, including K2 Thinking, and the `kimi-k2.5` endpoint are listed as discontinued.[30] The consumer product can apply lower limits based on the selected model, membership level, file handling, or other service constraints.[6] Because these limits and product plans can change without a new model release, they should be checked in the live documentation rather than inferred from historical launch announcements.

Kimi output is generated probabilistically. Research on [long-context language models](https://aiwiki.ai/wiki/long_context) has found that increasing the available context does not ensure that every part will be used equally well.[26] Search and agent modes likewise add tools without eliminating confabulation, [prompt injection](https://aiwiki.ai/wiki/prompt_injection), or unsafe tool actions, all of which are recognized risks for [generative AI](https://aiwiki.ai/wiki/generative_ai) systems.[27] High-impact uses require source review, least-privilege tool access, and independent verification.

## 3D Make

3D Make is a 3D creation feature in the Kimi assistant. Moonshot's help center describes it as powered by K3.1 Pro: a user describes an idea in natural language, or uploads a sketch or reference image, and Kimi produces an interactive 3D website or game, a 3D model, or a file intended for 3D printing.[37]

### Release

Moonshot's K3.1 Pro technical blog, published with the model on October 23, 2026, said that "3D Make is now live on kimi.ai and the Kimi app" and that users could preview and edit 3D models online, download them for 3D printing, or export them as videos or Live Photos. An Internet Archive capture of the blog made that day contains the passage.[28][42] ForkLog and WinBuzzer mentioned the feature in their launch coverage of K3.1 Pro.[39][40] On October 30, 2026, the official Kimi account posted "Introducing 3D Make in Kimi", saying that Kimi "connects to tools like Blender and CAD tools to create models, assemble scenes, and develop games" and that the feature was available on kimi.ai and in Kimi Work.[35] The October 23 blog named kimi.ai and the Kimi app as the places to use it; the October 30 post named kimi.ai and Kimi Work, the desktop agent. A Kimi Academy tutorial published alongside the announcement (its page shows an update date of October 29) says 3D Make was introduced "in Kimi.ai and Kimi Work".[36]

### Features and access

The help center's feature guide, as of October 30, 2026, documents three creation types, each of which can be chosen manually or inferred from the request:[37]

| Type | Documented output |
| --- | --- |
| 3D Web | Interactive 3D websites, web scenes, and web games, with online preview and sharing |
| 3D Modeling | 3D models generated from text, sketches, or reference images, which can be previewed and edited online |
| 3D Printing | A generated model exported as a print-ready file for use with slicing software |

Generated work can be rotated and zoomed in the browser, revised over several turns, exported as a video or Live Photo, and shared through a public link. The page also offers curated templates with a "Remix" option that reuses a template's original prompt. 3D Make is reached from a dedicated web page, from entries on the Kimi homepage, or in the Kimi mobile app for iOS, Android, and HarmonyOS. The guide says that generation and edits draw on Kimi's unified usage quota, that complex works may take a few minutes, and that the output is AI-generated and should be checked; for printed objects it advises attention to structural safety.[37]

A table in Moonshot's 3D tutorial lists the skills and plugins grouped under 3D Make and where each is available:[36]

| Skill or plugin | What Moonshot says it does | Where available |
| --- | --- | --- |
| game | Builds playable browser games, using Three.js for 3D by default | Kimi.ai and Kimi app; Kimi Work |
| content-creation | Creates 3D scenes, animation, SVG illustrations, voxel art, and Blender models | Kimi.ai and Kimi app; Kimi Work |
| 3d-printing | Creates 3D-printable models with Blender, CadQuery, and other tools | Kimi.ai and Kimi app |
| Tripo 3D Generation | Calls the [Tripo](https://aiwiki.ai/wiki/tripo) API to generate detailed GLB models of characters and animals | Kimi.ai and Kimi app; Kimi Work |
| 3D Editor | Imports assets into an editor for the agent to refine materials, lights, cameras, and animation, then exports a GLB file or video | Kimi Work |
| Build (webdev) | Creates and validates static sites, Three.js pages, and browser games, and publishes them as Kimi static websites | Kimi Work |

### Tools and modeling approach

Moonshot's tutorial divides the work into two methods. For parametric modeling of objects defined by dimensions and rules, such as mechanical parts, furniture, and buildings, Kimi writes code: JavaScript for Three.js, Python for Blender, and Python or C# for CadQuery, FreeCAD, and Rhino. For detailed meshes such as people, animals, statues, and complex props, it calls 3D generation services such as Tripo, Meshy, or Hyper3D. The tutorial says language models such as K3.1 Pro "still struggle" with mesh modeling, which is why that step is handed to diffusion-based generators.[36] Moonshot also says it trained K3.1 Pro to operate 3D tools through Python scripts, so that software without Model Context Protocol support can be used. Its launch blog says the model can script Blender and CAD software "without MCP connectors".[28][36] That is a different route from [Blender MCP](https://aiwiki.ai/wiki/blender_mcp) integrations, which connect agents to Blender through the protocol. The plugin set is mixed, however: the Autodesk Fusion plugin is listed as "Fusion 360 via MCP".[36]

For game development, the tutorial says Kimi has been adapted to Unreal Engine, Unity, and Godot.[36] On launch day, Moonshot said K3.1 Pro "works directly" in Blender, Unreal Engine, Godot, and CAD tools.[38] A screenshot of the 3D plugin category in the tutorial lists Tripo 3D generation, Blender, Unity, Unreal Engine, Rhino, Tripo AI, FreeCAD, Roblox Studio, 3D Slicer (described as local medical imaging scripts), and Autodesk Fusion, with a further link to Houdini, Autodesk Maya, and two more plugins. The tutorial states that plugins requiring locally installed software are available only in Kimi Work.[36] It recommends Kimi.ai for quick 3D websites and simple standalone models, Kimi Work or Kimi Code for local modeling software, game engines, and projects stored on the user's computer, and the Kimi API for integrating the models into other workflows.[36]

### Moonshot's tips

The tutorial's four worked examples are an Unreal Engine 5.8 scene with a MetaHuman character, a 3D-printable desk lamp exported as separate STL parts, a building modeled in Blender entirely through code, and a miniature city that combines Blender scripts with Tripo assets. Its five tips are:[36]

1. Supply reference images or videos. Moonshot suggests clear silhouettes and consistent multi-angle views, and dimensions such as hole diameter and plate thickness for mechanical parts.
2. Ask Kimi to turn references into a detailed production prompt, and check its assumptions before building.
3. Block out a scene with simple shapes to check scale and layout, then refine models, materials, and lighting, and attach screenshots when asking for fixes.
4. Start from existing asset and animation libraries. The tutorial names Poly Haven and Kenney, whose assets it says use the CC0 license, and the GSAP animation library.
5. Save a working workflow as a reusable skill.

The tutorial lists rigging, complex materials and physics simulation, and consistent large 3D worlds as limits Moonshot still wants to address, and says Kimi "can't guarantee a perfect result every time".[36]

### Independent coverage

ForkLog and WinBuzzer covered 3D Make within their reports on the K3.1 Pro launch. ForkLog reported that the Kimi apps had added a feature for viewing and editing generated models and exporting them for 3D printing. WinBuzzer reported that 3D Make lets users preview, edit, and export models in Kimi's web and mobile products.[39][40] The Chinese technology outlet Zhidongxi ran its own hands-on test of K3.1 Pro, not of 3D Make specifically. It reported that the model took about 45 minutes to produce an editable CAD model of a two-story exhibition hall with 3,052 components and 11 pages of drawings, and that Moonshot had launched a 3D model generation plugin in Kimi Work.[41]

## Weights and licensing

Moonshot describes K2-series releases as open models, but "open-weight" is the more precise term for checkpoints whose trained parameters can be downloaded. Availability of weights does not by itself provide the training data, complete preprocessing pipeline, or unrestricted rights to every use.

K2 is distributed under a modified [MIT license](https://aiwiki.ai/wiki/mit_license). In addition to the usual notice requirement, the license requires prominent Kimi K2 attribution in a product or service that has more than 100 million monthly active users or more than US$20 million in monthly revenue.[25] K2.5 and K2.6 repositories refer users to their included license files, so deployers should review the license attached to the exact checkpoint rather than assume that every Kimi model has identical terms.

K3 uses a separate Kimi K3 License. It permits use, modification, distribution, and derivative works subject to stated conditions. It also requires a separate agreement for certain model-as-a-service operators whose aggregate revenue exceeds US$20 million over a consecutive 12-month period, and it includes an attribution condition for products above specified user or monthly-revenue thresholds.[18] The license contains definitions, exceptions, and compliance obligations that cannot be reduced to a general claim that all commercial use is unrestricted.

## Evaluation and safety

Most launch-day capability claims for Kimi models come from Moonshot's own technical reports and model cards. The reports often describe evaluation settings, but benchmark results can still change with the prompt, sampling parameters, reasoning budget, tool availability, model revision, and harness. Scores from [SWE-bench Verified](https://aiwiki.ai/wiki/swe-bench_verified), [LiveCodeBench](https://aiwiki.ai/wiki/livecodebench), [AIME 2025](https://aiwiki.ai/wiki/aime_2025), or [GPQA](https://aiwiki.ai/wiki/gpqa) are not interchangeable, and a higher score on one does not establish general superiority.

An April 2026 preprint reported an independent safety evaluation of Kimi K2.5 across chemical, biological, radiological and nuclear information, cybersecurity, misalignment, political censorship, bias, and harmful-request handling. The authors found substantial dual-use capability and reported lower refusal rates on some hazardous prompts than comparison models. They also found no evidence of persistent malicious goals in the tested scenarios and described important limitations to their evaluation.[22] The study was a preliminary preprint, not a complete certification of either safety or danger.

In July 2026, the United Kingdom AI Security Institute and the United States Center for AI Standards and Innovation published a preliminary joint assessment of K3's cyber capabilities. On their selected tasks, K3 performed below the leading closed US models they tested but above another open-weight Chinese comparison model. The evaluators also reported that K3 attempted offensive cyber tasks when asked rather than consistently refusing them.[23] The assessment covered a limited task set, and the comparison models did not all use identical safeguards, so its findings should not be generalized beyond the tested conditions.

These results support a narrower conclusion than launch marketing: Kimi models are capable general-purpose systems with meaningful tool-use and dual-use potential, but their performance and safety depend on the model version, interface, evaluation setup, and deployment controls.

## See also

- [Moonshot AI](https://aiwiki.ai/wiki/moonshot_ai)
- [AI game development](https://aiwiki.ai/wiki/ai_game_development)
- [3D printing](https://aiwiki.ai/wiki/3d_printing)
- [Large language model](https://aiwiki.ai/wiki/large_language_model)
- [Multimodal AI](https://aiwiki.ai/wiki/multimodal_ai)
- [AI agent](https://aiwiki.ai/wiki/agent)
- [Mixture of experts](https://aiwiki.ai/wiki/mixture_of_experts)
- [Open-source artificial intelligence](https://aiwiki.ai/wiki/open_source_ai)
- [DeepSeek](https://aiwiki.ai/wiki/deepseek)
- [Qwen](https://aiwiki.ai/wiki/qwen)
- [Chain-of-thought prompting](https://aiwiki.ai/wiki/chain_of_thought)

## References

1. *Beijing Daily*, "A large-model startup founded by a post-1990 entrepreneur launches Kimi Chat" (October 11, 2023). https://news.bjd.com.cn/2023/10/11/10589091.shtml
2. AIbase, "Kimi Chat opens to the public and no longer requires beta qualification" (November 17, 2023). https://news.aibase.com/news/3269
3. *South China Morning Post*, "Alibaba-backed Moonshot AI claims breakthrough with expanded Chinese-character prompt for Kimi chatbot" (March 20, 2024). https://www.scmp.com/tech/big-tech/article/3256109/alibaba-backed-moonshot-ai-claims-breakthrough-expanded-chinese-character-prompt-kimi-chatbot
4. Qin et al., "Mooncake: A KVCache-centric Disaggregated Architecture for LLM Serving," arXiv:2407.00079 (2024). https://arxiv.org/abs/2407.00079
5. Kimi Help Center, "Kimi overview," accessed July 28, 2026. https://www.kimi.com/help/getting-started/overview
6. Kimi Help Center, "Capabilities," accessed July 28, 2026. https://www.kimi.com/help/others/capability
7. Kimi Help Center, "Kimi API troubleshooting," accessed July 28, 2026. https://www.kimi.com/help/kimi-api/api-troubleshooting
8. Kimi Team et al., "Kimi k1.5: Scaling Reinforcement Learning with LLMs," arXiv:2501.12599 (2025). https://arxiv.org/abs/2501.12599
9. Kimi Team et al., "Kimi K2: Open Agentic Intelligence," arXiv:2507.20534 (2025). https://arxiv.org/abs/2507.20534
10. Moonshot AI, "Kimi K2" source repository and model documentation, accessed July 28, 2026. https://github.com/MoonshotAI/Kimi-K2
11. Moonshot AI, "Kimi-K2-Instruct-0905" model card, accessed July 28, 2026. https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905
12. Moonshot AI, "Kimi K2 Thinking," November 2025. https://www.kimi.com/blog/kimi-k2-thinking
13. Kimi Team et al., "Kimi K2.5: Visual Agentic Intelligence," arXiv:2602.02276 (2026). https://arxiv.org/abs/2602.02276
14. Moonshot AI, "Kimi K2.5" source repository and model documentation, accessed July 28, 2026. https://github.com/MoonshotAI/Kimi-K2.5
15. Moonshot AI, "Kimi K2.6: Intelligence in Motion," April 2026. https://www.kimi.com/blog/kimi-k2-6
16. Moonshot AI, "Kimi-K2.6" model card, accessed July 28, 2026. https://huggingface.co/moonshotai/Kimi-K2.6
17. Moonshot AI, "Kimi K3" source repository and model card, accessed July 28, 2026. https://github.com/MoonshotAI/Kimi-K3
18. Moonshot AI, "Kimi K3 License," July 2026. https://github.com/MoonshotAI/Kimi-K3/blob/main/LICENSE
19. Kimi Help Center, "Kimi Agent overview," accessed July 28, 2026. https://www.kimi.com/help/agent/agent-overview
20. Kimi Help Center, "Deep Research overview," accessed July 28, 2026. https://www.kimi.com/help/deep-research/deep-research-overview
21. Associated Press, "China's new AI model halts new subscriptions as demand swamps capacity" (July 20, 2026). https://apnews.com/article/4c66a2e0f557ce79d3cc2d769c9a6226
22. Yong et al., "An Independent Safety Evaluation of Kimi K2.5," arXiv:2604.03121 (2026). https://arxiv.org/abs/2604.03121
23. UK AI Security Institute, "Preliminary assessment of Kimi K3's cyber capabilities" (July 2026). https://www.aisi.gov.uk/blog/preliminary-assessment-of-kimi-k3s-cyber-capabilities
24. Kimi Help Center, "Kimi API overview," accessed July 28, 2026. https://www.kimi.com/help/kimi-api/api-overview
25. Moonshot AI, "Kimi K2 License," accessed July 28, 2026. https://github.com/MoonshotAI/Kimi-K2/blob/main/LICENSE
26. Liu et al., "Lost in the Middle: How Language Models Use Long Contexts," *Transactions of the Association for Computational Linguistics* 12 (2024). https://aclanthology.org/2024.tacl-1.9/
27. National Institute of Standards and Technology, "Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile," NIST AI 600-1 (2024). https://doi.org/10.6028/NIST.AI.600-1
28. Moonshot AI, "Kimi K3.1 Pro Tech Blog: Advancing Open Intelligence," kimi.ai, accessed October 25, 2026. https://www.kimi.ai/blog/kimi-k3-1-pro
29. Kimi.ai (@Kimi_Moonshot), "Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model," X (October 23, 2026). https://x.com/Kimi_Moonshot/status/2113680701770068067
30. Kimi API Platform, "Model List," accessed October 30, 2026. https://platform.kimi.ai/docs/models
31. Kimi Code Docs, "What's New" (changelog entries from May to October 2026), accessed October 30, 2026. https://www.kimi.com/code/docs/en/kimi-code/whats-new.html
32. Moonshot AI, "Kimi-K2.7-Code" model card, Hugging Face, accessed October 30, 2026. https://huggingface.co/moonshotai/Kimi-K2.7-Code
33. Kimi Help Center, "What Is Kimi Work?," accessed October 30, 2026. https://www.kimi.com/en/help/kimi-work/overview
34. Kimi Help Center, "Kimi Work Release Notes," accessed October 30, 2026. https://www.kimi.com/en/help/kimi-work/release-notes
35. Kimi.ai (@Kimi_Moonshot), "Introducing 3D Make in Kimi," X (October 30, 2026). https://x.com/Kimi_Moonshot/status/2116005725814280570
36. Kimi Academy, "How to Create Better 3D Projects with Kimi," updated October 29, 2026, accessed October 30, 2026. https://www.kimi.ai/academy/3d-tutorial
37. Kimi Help Center, "Kimi 3D Make: Feature Guide," accessed October 30, 2026. https://www.kimi.com/en/help/3d-make/3d-make-overview
38. Kimi.ai (@Kimi_Moonshot), "K3.1 Pro works directly in the tools used by the 3D modeling and game industries," X (October 23, 2026). https://x.com/Kimi_Moonshot/status/2113680710150357279
39. ForkLog, "Moonshot AI launches Kimi K3.1 Pro, restricts its cyberattack capabilities" (October 24, 2026). https://forklog.com/en/moonshot-ai-launches-kimi-k3-1-pro-restricts-its-cyberattack-capabilities/
40. Markus Kasanmascheff, "New Chinese Kimi K3.1 Pro AI Model Outperforms GPT-6 Astra With Near Claude Opus 5.5 Performance," WinBuzzer (October 25, 2026). https://winbuzzer.com/2026/10/25/kimi-k3-1-pro-gains-ground-in-ai-coding-tests-xcxwbn/
41. Zhidongxi, hands-on report on Kimi K3.1 Pro (in Chinese), republished by Anue (cnyes.com) (October 24, 2026). https://hao.cnyes.com/post/274052
42. Internet Archive, capture of "Kimi K3.1 Pro Tech Blog: Advancing Open Intelligence" (October 23, 2026, 17:12 UTC). https://web.archive.org/web/20261023171225/https://www.kimi.ai/blog/kimi-k3-1-pro
43. Kimi Help Center, "What can Kimi do?," accessed October 30, 2026. https://www.kimi.com/en/help/others/capability
44. Kimi Help Center, "What are the differences between Kimi, Kimi Work, Kimi Code, and Kimi Claw? Which one should I use?," accessed October 30, 2026. https://www.kimi.com/en/help/others/product-comparison

