DeepSeek — “They want you” to train their models.

People are using AI for free, and it's not at all what you think. You can use top models without even providing your credit card information. I'm talking about API keys. I'm talking about spinning up agents, doing all kinds of cool stuff, and it's because of a very unlikely source. Perhaps until you think about it. I'm Gagatjian, and today I'm gonna tell you how to use NIMM console to do some crazy stuff. NIMM stands for NVIDIA inference microservices. It's basically like Docker for AI models. That means it's a ready-made set of containerized software modules to basically let you set up AI models anywhere, and it does more than that. First, let's talk about the models because there are a lot to play with. You've got Mistral Large 3, which is a beast at multilingual reasoning and planning. You've got Kimi 2.6, that Kimi 2.6, and MOE beast with over a trillion parameters. This is your supercoder that you can use for free. You've got GLM 5.1 with advanced multi-turn reasoning. This is your workhorse for long context analysis. You've got Nemo Tron Nano Omni for native audio and video processing. You've got Deepseek, you've got Gemma, the list goes on. And the catch you're waiting for is basically non-existent. Yes, there's a rate limit, but it's generally 40 requests per minute. There are additional hidden limits if you try to go crazy with really expensive models like 2.6. That one's roughly 30 requests per hour, but that shouldn't be a problem unless you're doing something…