There is a new language model developed by the Chinese, a free one developed by Moonshot AI. The interface is in Chinese, but logically clear. Understands Russian without any problems. For us, humanitarians, the following features can be useful:

1. Work with three-dimensional cases. The model accepts up to 200,000 tokens (approximately 1,500 pages of text), which allows you to upload a complete collection of works, a series of letters, or a set of articles at a time and immediately ask a question about the entire array. At the same time, links to specific pages and quotes are saved.

2. Support for multilingual sources. Kimi reads and collates materials in dozens of languages, including rare and historical ones. This is convenient when working with documents in Latin, classical Chinese, Arabic, or Hebrew without prior translation.

3. Structured data extraction. From a large array of text, you can automatically get: - a chronology of events; - a list of the mentioned persons with biographical notes; - a list of terms and concepts with contexts of use. The result is displayed in a table or in CSV format, ready for import into Excel or a database.

4. Comparative analysis of sources. The model is able to compare the contents of several manuscripts or text revisions and indicate points of divergence: added or omitted paragraphs, changes in wording, shifts in emphasis.

5. Preparation of citations and bibliographies. Upon request, excerpts of the required length are generated with precise page links; design styles (APA, MLA, Chicago) are set in one line.6. Long-term discussion. In a single dialog, you can return to previously uploaded content hours or days later: the model retains the context and continues reasoning without re-uploading files.

Conclusion. For humanitarian projects, where necessary:• upload 1,000-1,500 pages of text in different languages at once,• get accurate quotations without losing pages,• compare editions or translations without manual splitting into parts, Kimi remains the only language model that performs these operations in a single query and without additional tools. GPT-4o and Claude 4 are superior in image tasks, and Gemini 2.5 is superior in volume, but as the context grows, the detail accuracy decreases faster than that of Kimi.