Coding
Moonshot AI reports K3 can sustain long-horizon software engineering sessions, navigating large codebases and orchestrating terminal tools with minimal human oversight. [S1]
Independent Kimi K3 guide
Kimi K3: An Independent Resource Guide
Kimi K3 Quick Experience
Ask one question about Kimi K3.
What Kimi K3 Supports
Capabilities described in official sources
K3 supports long-context coding, multimodal input, tool calling, and research workflows. [S1, S2]
CodingMultimodal InputTool-Supported WorkflowsResearch and Knowledge Work
Multimodal Input
Moonshot AI describes K3 as natively supporting visual understanding. It accepts text, image, and video as input. [S1, S2]
Tool-Supported Workflows
K3 supports tool calling and is compatible with agentic workflows. It can orchestrate multiple tools within a single session. [S2]
Research and Knowledge Work
Moonshot AI reports that K3 produces interactive research reports, data visualizations, dashboards, and publication-quality materials. [S1]
All capability descriptions below are based on official sources. K3 accepts text, images (suggested <= 4K resolution), and video (suggested <= 1080p). Its 1,048,576-token context window supports very long inputs, and automatic context caching activates when prompts exceed 256 tokens. Moonshot AI also reports benchmark results across coding, productivity, and multimodal evaluations. All results are vendor-reported. See the official Kimi K3 blog for complete benchmark charts. [S1, S2]
Choose your next step
Find the Kimi K3 guide you need
Start with one focused page instead of making every technical topic compete on the homepage.
01
Kimi K3 overview
Core model positioning, official capabilities, and verified specifications.
Read guide 02Kimi K3 open weights
Audited repository files, license conditions, and pinned download notes.
Read guide 03Kimi Code
Terminal coding workflows, model selection, and official setup references.
Read guide 04Kimi K2.7 Code
A focused guide to the coding model and its documented relationship to K3.
Read guide 05K3 vs K2.7 Code
A source-checked side-by-side comparison of the two model entries.
Read guide 06Kimi K3 blog
Research notes, API walkthroughs, and dated release updates.
Read guideThird-Party Benchmark Snapshot
Third-Party Benchmark Snapshot
BenchLM BenchAlign v5.2 (July 23, 2026): Kimi K3 ranked #4 of 200 with a score of 80.96. Confidence: Low.
Rank #4 of 200 models. Overall score 80.96. Confidence Low. 57 of 323 tracked benchmarks were published for Kimi K3.
Read BenchLM methodology and third-party sources
In BenchLM's July 23, 2026 BenchAlign v5.2 snapshot, Kimi K3 ranked #4 of 200 models with an overall score of 80.96. BenchLM assigns Low confidence: of 323 tracked benchmarks, 57 were published for K3, and three of eight evaluation categories — Math, Multilingual, and Instruction Following — had no data. The ranking is a third-party composite; BenchLM aggregates and normalizes vendor-reported scores. It does not independently test models. BenchLM is a third-party ranking website. This composite ranking is produced by BenchLM and is not an official Moonshot AI statement.
Third-Party Sources
- T1BenchLM — Kimi K3 Model ProfileAccessed July 24, 2026
- T2BenchLM — MethodologyAccessed July 24, 2026
- T3BenchLM — Leaderboard HomepageAccessed July 24, 2026
Users say
Public Kimi K3 user feedback
Short excerpts from publicly accessible Jike, X, and Reddit posts. These are personal experiences, not a unified test result or an OpenK3 conclusion.
“Kimi K3 recreated a complex motion-video effect with high fidelity.”
A public user example focused on visual recreation.
“Strong results on a HyperFrames benchmark, with only a few prompts needed to recreate the effect.”
A public post describing one benchmark-style workflow.
“Creative writing felt more able to follow a personal tone and voice.”
An individual experience from a public discussion.
“High-level reasoning, engineering detail, and abstract thinking felt strong.”
A positive individual report from the same public discussion.
Current Status
Availability, weights, license, and verification date
Available today via the Kimi API
The model ID currently listed in official documentation is kimi-k3. It is also accessible through the Kimi.com web interface, Kimi Work desktop app, and Kimi Code terminal agent. [S3, S2, S1]
Open weights released July 27, 2026
The official moonshotai/Kimi-K3 Hugging Face repository is public and contains 118 files, including 96 Safetensors weight shards and 4 vendor-reported evaluation metadata YAML files. [S19]
Custom Kimi K3 License
The open-weight release uses the custom Kimi K3 License. Commercial use is permitted subject to its conditions; read the full license before use. [S21]
July 27, 2026
Open-weight facts were verified against Hugging Face revision 9f62e4e9fffbd0a83ddd60e1c209d828994b3569. [S19]
For the pinned repository revision, file sizes, download commands, and license conditions, read the Kimi K3 Open Weights guide.
Kimi K3 Specifications
Key specs at a glance
View the full official specification table
| Specification | Value |
|---|---|
| Developer | Moonshot AI [S1, S8, S10] |
| Architecture | Mixture of Experts with KDA and Attention Residuals (AttnRes) [S1] |
| Total parameters | 2.8 trillion [S1, S2] |
| Active parameters | 104 billion [S19] |
| Experts | 896 total / 16 activated per token [S1] |
| Context length | 1,048,576 tokens (1M) [S1, S2, S4] |
| Input modalities | Text, image (PNG/JPEG via base64), video (via file upload) [S1, S2] |
| Output modality | Text [S2] |
| API model ID | kimi-k3 [S2, S3] |
| License | Kimi K3 License (custom) [S21] |
K3 always operates in thinking mode; reasoning effort can be set to low, high, or max (default). Moonshot AI reports approximately 2.5x better scaling efficiency compared to Kimi K2. [S2, S4, S1]
FAQ
Common questions about Kimi K3
What is Kimi K3?
Kimi K3 is Moonshot AI's flagship large language model, introduced in July 2026. It is a 2.8-trillion-parameter Mixture-of-Experts model with a 1,048,576-token context window. It accepts text, images, and video as input and is designed for long-horizon coding, knowledge work, and deep reasoning. The model is available via the Kimi API, and its open weights were released on Hugging Face on July 27, 2026. [S1, S2, S19]
Is Kimi K3 available now?
Yes. Kimi K3 is accessible through Kimi.com, Kimi Work, Kimi Code, and the Kimi API with model ID kimi-k3. Its open weights are also publicly available from the official moonshotai/Kimi-K3 Hugging Face repository. [S1, S2, S3, S19]
What license covers the Kimi K3 open weights?
Kimi K3 uses the custom Kimi K3 License, which has no standard SPDX identifier. It permits commercial use subject to conditions, including requirements for certain Model-as-a-Service businesses and very large commercial products. Read the full license and the Open Weights guide before use. [S21]
What is the Kimi K3 context window?
Kimi K3 supports a context window of 1,048,576 tokens, or 1 million tokens. The default output token limit is 131,072, configurable up to the full 1,048,576. Automatic context caching is built in. [S1, S2, S4]
How much does the Kimi K3 API cost?
As of July 24, 2026, official API pricing is ¥2.00 cached input, ¥20.00 uncached input, and ¥100.00 output per 1M tokens in CNY, and approximately $0.30, $3.00, and $15.00 per 1M tokens in USD, as reported by Moonshot AI. Prices may change. For current pricing, rate limits, and setup instructions, check the official Kimi API documentation. [S5, S1, S2]
Does Kimi K3 replace Kimi K2.7 Code?
The official sources reviewed for this page did not state that Kimi K3 replaces Kimi K2.7 Code. As of July 24, 2026, both kimi-k3 and kimi-k2.7-code appeared in the official Kimi API model list. For a side-by-side spec comparison, see the K3 vs K2.7 Code page. [S3]
Official Sources
Source documents used for this page
Official Kimi, Moonshot AI, and repository sources are listed below with access dates.
View all official sources (13)
- [S1]Kimi K3: Open Frontier Intelligence
Moonshot AI official blog, published July 17, 2026
Accessed July 24, 2026https://www.kimi.com/blog/kimi-k3 - [S2]Kimi K3 Quickstart Guide
Kimi API documentation
Accessed July 24, 2026https://platform.kimi.com/docs/guide/kimi-k3-quickstart - [S3]
- [S4]API Parameter Reference
Kimi API documentation
Accessed July 24, 2026https://platform.kimi.com/docs/api/models-overview - [S5]Kimi K3 Pricing (CNY)
Kimi API documentation
Accessed July 24, 2026https://platform.kimi.com/docs/pricing/chat-k3 - [S8]Moonshot AI GitHub
Official Moonshot AI GitHub organization
Accessed July 24, 2026https://github.com/MoonshotAI - [S9]Moonshot AI Hugging Face
Official Moonshot AI Hugging Face profile
Accessed July 24, 2026https://huggingface.co/moonshotai - [S10]Moonshot AI Corporate Website
Moonshot AI corporate website
Accessed July 24, 2026https://www.moonshot.ai/ - [S17]Kimi Terms of Service
Kimi terms of service, effective January 21, 2026
Accessed July 24, 2026https://www.kimi.com/user/agreement/modelUse?version=v2 - [S18]Kimi-K2 LICENSE
Modified MIT License for K2 predecessor only; does not indicate K3 license
Accessed July 24, 2026https://github.com/MoonshotAI/Kimi-K2/blob/main/LICENSE - [S19]Moonshot AI — Kimi K3 Hugging Face Model Repository
Official Kimi K3 open-weight repository
Accessed July 27, 2026 at 16:45 UTChttps://huggingface.co/moonshotai/Kimi-K3 - [S21]Kimi K3 License
Custom license for Kimi K3 code and model weights, pinned to the audited revision
Accessed July 27, 2026https://huggingface.co/moonshotai/Kimi-K3/blob/9f62e4e9fffbd0a83ddd60e1c209d828994b3569/LICENSE - [S22]MoonshotAI/Kimi-K3 Technical Report Repository
Official technical report repository
Accessed July 27, 2026https://github.com/MoonshotAI/Kimi-K3
Latest guides
Recent Kimi K3 notes and walkthroughs
Browse the full archive in the OpenK3 blog.
About OpenK3
an independent, source-checked Kimi K3 reference site
OpenK3 is an independent, source-checked Kimi K3 reference site. We are not affiliated with Moonshot AI. Every factual claim on this page is linked to a specific official source above.
Confirmed
Directly stated in an official source.
Not publicly disclosed
Information that has not been released by Moonshot AI.
Time-sensitive
Repository revisions, license terms, pricing, and access conditions require dated checks.
We do not publish search volume estimates, model parameter counts that sources do not provide, unverified benchmark conclusions, or statements about which model is best. All benchmark mentions are attributed as vendor-reported. The comparison page at /kimi-k3-vs-kimi-k2-7-code provides a side-by-side official-specifications breakdown of K3 and K2.7 Code. The /kimi-k3-open-weights guide documents the audited repository revision, download commands, and license conditions. Learn more about the site on the About page or contact us.