AI Pulse · AI Trends Pulse

Moonshot AI just released Kimi K3, and the numbers are wild. It’s a 2.8 trillion parameter mixture-of-experts model, with 104 billion parameters active at any given time. It handles a million-token context window and includes native vision capabilities. That makes it the largest open-weights model anyone’s shipped so far.

The timing matters. Together AI and Modal both added hosting support on day one, so developers can spin it up without building their own infrastructure. The model’s been trending hard across X, Hacker News, and Hugging Face this week, and it’s feeding a narrative that’s hard to ignore: open models are closing the gap with proprietary ones, and they’re getting cheaper to run.

For business owners, this is another data point in a pattern. Six months ago, if you wanted frontier-level AI performance, you paid OpenAI or Anthropic. Now you’ve got open-weights alternatives that can handle serious workloads, run on your own terms, and cost less per token. That changes the economics of building AI into your operations. You’re not locked into one vendor’s pricing or usage policies. You can test, iterate, and deploy without waiting for API rate limits or worrying about data leaving your environment.

This is the kind of shift we track closely when we build tools like the Omni Command Centre, where you need models that can handle large documents, complex reasoning, and multi-modal inputs without breaking the budget or your compliance posture. Kimi K3 won’t be the right fit for every use case, but it expands the menu of what’s possible when you’re designing systems that need to scale. The full breakdown has more technical detail if you want to dig in.