What is Kimi K3? It is Moonshot AI’s flagship open-weight multimodal AI model, built for long-context reasoning, coding, knowledge work and agentic tasks. It uses a 2.8-trillion-parameter sparse Mixture-of-Experts architecture, supports a 1-million-token context window and activates about 104 billion parameters per token. Moonshot AI released Kimi K3 on July 16, 2026, and published the full model weights on July 27, 2026.
Table of Contents
What Is Kimi K3?
Kimi K3 is the flagship model from Moonshot AI, the company behind the Kimi family of AI products. It is an open-weight, natively multimodal model designed for demanding work such as software engineering, long-form reasoning and research, where an AI system has to work through large amounts of information.
Kimi K3 has 2.8 trillion total parameters. Unlike a conventional dense model, which uses the whole network for every token, it uses a sparse Mixture-of-Experts design. Technical summaries describe it as selecting 16 of 896 experts for each token, which means only a fraction of the total parameters are active at once.
When Did Kimi K3 Come Out?
When did Kimi K3 come out? Moonshot AI announced and released it on July 16, 2026, initially through Kimi’s apps and API. Coverage of the open-weight release says the full weights were published on Hugging Face on July 27, 2026, delivering a promise made at launch.
The gap between the two dates matters. Developers could use the hosted model from July 16, but they could only self-host it after the weights were published on July 27.
Key Features of Kimi K3
2.8 trillion parameters. Parameters are the learned values inside a model. A large count indicates scale, but it does not automatically make a model better.
Sparse Mixture-of-Experts architecture. The model routes each token to a small group of specialist experts. The idea is similar to a large organisation that assigns only the relevant teams to a given task. The approach aims to keep a very large capacity while limiting the computation needed for each token.
A 1-million-token context window. Kimi K3 supports 1,048,576 tokens of context. That allows a single task to include very large codebases, lengthy documents or extended research material. A large window does not guarantee equally strong results at every length, and performance depends on the task and how the information is organised.
Native multimodal capabilities. Kimi K3 accepts image input alongside text, and Moonshot positions it for multimodal knowledge work and agentic scenarios.
Kimi Delta Attention. For a closer look at what is Kimi K3 under the hood, Moonshot says its Kimi Delta Attention mechanism and its sparse MoE design support better scaling efficiency than the previous Kimi K2 generation, so the model can deliver more capability for the computing resources used.
What Can Kimi K3 Be Used For?
Kimi K3 is positioned for tasks that go beyond simple question answering:
- Software development, code generation, code review and work across large repositories
- Long-form research and analysis involving large volumes of information
- Reasoning and problem-solving
- Working with images alongside text
- Agentic workflows that use tools across several steps
- Producing and transforming complex knowledge-work outputs
Is Kimi K3 Open Source or Open Weight?
What is Kimi K3 in licensing terms? It is more accurately described as open-weight. Its weights were published on July 27, 2026, and reporting on the release identifies the license as a Modified MIT license. Sources differ slightly on the license name, so check the license file in the official repository before redistributing the model.
The terms “open source” and “open weight” are often used interchangeably, but they are not identical. Open weight means the trained parameters are publicly available, while the licence sets the conditions for use and redistribution.
Kimi K3 Pricing: How Much Does It Cost?
What is Kimi K3 going to cost to use? Its API uses usage-based pricing. Reported Moonshot API rates are:
| Usage | Price per 1M tokens | What it means |
|---|---|---|
| Cached input | $0.30 | Input tokens that were previously cached |
| Input | $3.00 | New input tokens sent to the model |
| Output | $15.00 | Tokens generated by the model |
These are API rates. They are not the cost of running the open-weight model yourself. Self-hosting requires hardware, memory, electricity, software and operations, and the published model files are very large.
Why Did Kimi K3 Affect AI and Chip Stocks?
Kimi K3 was launched in the same week that chip stocks were already under pressure, and the connection is debated. Some reports tie the sell-off directly to the launch. Others point to wider concerns about AI infrastructure spending and a report of a Chinese chip breakthrough.
Reports from the week describe a sharp fall in Asian chip stocks. AP reported that Tokyo’s Nikkei 225 fell 4% and Taiwan’s Taiex skidded 4.7% on Tuesday, 28 July 2026. Kimi K3 was among several developments that revived a question raised during the DeepSeek episode: if powerful models can be built and served more efficiently, will the industry need as much expensive computing infrastructure as investors expected?
It would be misleading to say Kimi K3 alone determines future demand for AI chips. A large, popular model also creates demand for computing capacity, so efficiency gains do not necessarily reduce infrastructure spending.
Why Kimi K3 Matters
What is Kimi K3’s real significance? It is not simply its parameter count. Its combination of very large total capacity, sparse expert activation, long context, multimodal input and open weights shows that frontier AI development is increasingly focused on efficiency and deployability as well as raw scale.
For developers and organisations, open weights offer more control over deployment and experimentation than a hosted-only model. The size and operational needs of K3 mean that self-hosting is not a lightweight project, though.
Frequently Asked Questions About Kimi K3
What is Kimi K3? Kimi K3 is Moonshot AI’s 2.8-trillion-parameter open-weight, natively multimodal model. It uses a sparse Mixture-of-Experts architecture, supports a 1-million-token context window and is designed for reasoning, coding, knowledge work and agentic tasks.
When were the Kimi K3 model weights released? The full weights were released on July 27, 2026. Moonshot AI launched the model on July 16.
When did Kimi K3 come out? Kimi K3 launched on July 16, 2026. Its open-weight release followed on July 27, 2026.
How does Kimi K3 pricing work? Kimi K3 uses usage-based API pricing. Reported rates are $0.30 per million cached input tokens, $3.00 per million input tokens and $15.00 per million output tokens. Self-hosting is a separate cost.
Conclusion
What is Kimi K3 in short? It is a major open-weight release from Moonshot AI. It combines a 2.8-trillion-parameter sparse Mixture-of-Experts design with a 1-million-token context window, multimodal input and a focus on long-horizon reasoning and agentic work. Its July 2026 release also shows that the AI race is increasingly about how efficiently powerful models can be trained, served and deployed, not only how large they are.
Also read: GPT-6 Astra: What Makes OpenAI’s New AI Model So Powerful?
