@alexdyor
🛎 Chinese company Moonshot AI has released an open multimodal model Kimi-VL
Chinese company Moonshot AI has developed a new open artificial intelligence model Kimi-VL with 2.8 billion active parameters. The model processes text, images and videos, has a context window of 128,000 tokens.
Kimi-VL uses an "expert mixture" architecture, activating only the relevant part of the model for each task. According to the developer, the model outperforms its counterparts in 19 out of 24 test indicators.
Functionality includes analysis of full screenshots, handwriting recognition and solving mathematical problems in images. The model also interprets graphical interfaces and automates digital tasks.
The improved version of Kimi-VL-Thinking is focused on complex logical operations. The demo version is available at https://huggingface.co/spaces/moonshotai/Kimi-VL-A3B-Thinking, the related model Kimi k1.5 is presented at https://kimi.ai.
0 reply
0 recast
0 reaction