Every figure, licence term and definition stated in this video, chased to the place it was published. Checked 8 August 2026.
Moonshot describes Kimi K3 as an open-weight model, not an open-source one. The model card opens by calling it an “open-weight, native multimodal agentic model”, and says the full model weights are released under the Kimi K3 License.
The architecture figures.
| Total parameters | 2.8 trillion |
| Active parameters per token | 104 billion |
| Routing | 16 of 896 experts per token |
| Context window | 1,048,576 tokens |
| Weights published | 27 July 2026 |
Source: the Kimi K3 model card, https://huggingface.co/moonshotai/Kimi-K3
Kimi K2.5 was described as open-source. Its model card calls it “an open-source, native multimodal agentic model”, and states that “both the code repository and the model weights are released under the Modified MIT License.”
That is the change the video turns on: the same lab, one release apart, moved from “open-source” to “open-weight” in its own words.
For most of its length the licence is permissive. It grants, free of charge, the rights to use, copy, modify, publish, distribute, sublicense and sell the software, and to deploy, run, fine-tune and build derivative works from the model weights.
Two conditions sit on top of that.
The Model-as-a-Service threshold. A licensee running a Model-as-a-Service business, meaning giving third parties inference or fine-tuning access with meaningful control over inputs, parameters or training data, must reach a separate agreement with Moonshot once revenue across the licensee and its affiliates passes 20 million US dollars over any consecutive twelve month period.
The attribution threshold. A commercial product or service using Kimi K3 that exceeds 100 million monthly active users, or 20 million US dollars in monthly revenue, must display “Kimi K3” prominently in its interface, subject to the licence’s own exceptions. Internal use, and use through Moonshot’s official products or certified inference partners, are exempt.
The four freedoms. Version 1.0 of the Open Source AI Definition says an Open Source AI system must grant the freedoms to:
The first of those is the one that does the work in this video: for any purpose and without having to ask for permission.
The preferred form to make modifications. The definition requires all three of:
Weights are one of the three, which is the distinction the video is built on.
What Data Information has to cover. The definition asks for the complete description of all data used for training, including unshareable data, “disclosing the provenance of the data, its scope and characteristics, how the data was obtained and selected, the labeling procedures, and data processing and filtering methodologies”, plus a listing of publicly available training data and where to obtain it, and a listing of training data obtainable from third parties, including for a fee.
It does not require every individual training sample to be downloadable, which is why datasets containing copyrighted, licensed, private or otherwise non-redistributable material do not automatically put a system outside the definition.
Linux publishes the source used to build the kernel, not only the built kernel, and it is distributed under the GNU General Public License version 2, which grants the rights to inspect, modify, compile and redistribute.
Apache License 2.0, named in the video as the permissive end of the model licensing range, is an OSI-approved licence with no commercial-use threshold.
Reuters, 7 August 2026. Alibaba plans to require large commercial users of the open-weight version of its next Qwen model to share a portion of the revenue they generate from it, with the measure intended to arrive alongside the open-weight release. The rate was reported as not yet finalised.
The report draws the comparison to Moonshot directly: Moonshot’s terms require any party selling Kimi K3 as a service and generating more than 20 million dollars in annual revenue to reach a commercial agreement, which can include a revenue share of up to 30 per cent.
Small developers and researchers are not affected by the reported change.