Kimi K2.5 is an open-source, native multimodal agentic model built through continual pretraining on approximately 15 trillion mixed visual and text tokens atop Kimi-K2-Base.
It seamlessly integrates vision and language understanding with advanced agentic capabilities, instant and thinking modes, as well as conversational and agentic paradigms.
00:00 - Kimi K2.5 intro
00:30 - Kimi K2.5 specs
01:21 - Eye puzzle test
03:18 - Insane maze test
04:08 - Android os
06:51 - Hand tracking game
08:20 - Agent swarm tests
11:01 - Slides, sheet, doc agent test
13:48 - Master AI agents playbook
14:48 - Figma clone
16:09 - Video to website
16:55 - Trello clone
17:45 - Finding Waldo
18:07 - Deep medical research
21:06 - Kimi K2.5 specs
21:55 -'Benchmarks and cost
24:11 - How to use Kimi K2.5
25:18 - The value of open source