Minimax M1 — like James Bond among virtual agents.
(This is not a post about video generation, just to clarify.)
The online resource reports that our favorite Minimax has released a new open model — M1.
What’s interesting about it? It can process up to a million tokens in input (yes, that’s a lot) and generate output up to 80 thousand tokens.
Based on various benchmarks, M1 performs as well as, or even better than, giants like Gemini 2.5 Pro, DeepSeek-R1, Qwen3-235B, and others.
The main point here is that the model is open — you can download it via this link: https://github.com/MiniMax-AI/MiniMax-M1
But the most remarkable thing is that you get a powerful Manu with an enormous context, capable of doing everything. It feels like you have an agent that can do it all!
I decided to test it on the fly and came up with this simple prompt:
generate short movie about little cat. Generate it using your video generation capability, combine it with the story text, and present the entire thing as a webpage.
It took about 10 minutes — watching how Linux commands and the bot’s reasoning work was an extra pleasure. In the end, it generated for me this page:
https://vbzj9owyu5.space.minimax.io/ — scroll down and don’t focus only on the video, the main thing is to appreciate the fact of what’s happening.
Note: the video was created without a watermark, even though in the logs I saw the message: “Now I’ll generate the video using the minimax video generation tool.”
Logs and chat history can be found here: https://agent.minimax.io/chat?id=280996799635639
And don’t get carried away with video or image generation with logos like alamy — the most important thing is to evaluate the system’s performance.
It’s still unclear how long Minimax servers will be running, but it seems like we’re witnessing some kind of universal agent miracle — with a charming context window of a million tokens in input.
Try it yourself at https://agent.minimax.io/ — everything is free for now.