Kimi K3 FAQ: Context, Vision, Access and Limits
Kimi is the product family; K3 is the flagship model powering new agent experiences across chat, knowledge work, coding and the API. This page separates confirmed launch facts from details that still depend on a product plan or future weight release.
Kimi and K3 in one minute
Kimi.com is the general assistant, Kimi Work focuses on knowledge-work deliverables, Kimi Code runs agentic coding workflows, and Kimi API brings the model into developer products. K3 is available across those surfaces, though model IDs and plan limits differ.
Confirmed launch facts
K3 has 2.8T total parameters, native vision and an advertised context window up to one million tokens. It uses a sparse mixture of experts with 16 of 896 experts active and is designed for long-horizon coding, knowledge work and reasoning.
Where expectations need caution
One million tokens is a maximum, not a recommendation to send unfiltered data. Public benchmarks depend on their harnesses. Open-model language does not remove the need to check the final weight license. And Kimi itself documents sensitivity to thinking history, excessive proactiveness and a user-experience gap versus the strongest proprietary models.
Frequently asked questions
What is Kimi K3?
Kimi K3 is Moonshot AI's 2.8T-parameter flagship model for long-horizon coding, knowledge work, reasoning and native visual understanding.
How long is the Kimi K3 context window?
The model supports up to one million tokens. Actual product access can depend on the client and membership plan.
Can Kimi K3 understand images and video?
K3 is natively multimodal and the launch materials describe text, image and video understanding within the same model.
Where can I use Kimi K3?
It launched on Kimi.com, Kimi Work, Kimi Code and the Kimi API.
Is Kimi K3 open source?
Moonshot calls it an open 3T-class model and scheduled full weights by July 27, 2026. The final repository and license should be checked when released.
What are K3's known limitations?
Officially documented limitations include sensitivity to incomplete thinking history, overly proactive behavior on ambiguous tasks and a remaining UX gap versus the strongest proprietary models.
Primary sources
This guide prioritizes official Kimi and Moonshot AI materials. Product access and pricing can change, so confirm live details at the linked sources.