AI Pulse · AI Trends Pulse

A Chinese lab called Moonshot AI just dropped Kimi K3, a 2.8 trillion parameter model with a one million token context window. It’s open-weight, meaning you can download and run it yourself. The pricing is aggressive: about $0.70 per million input tokens and $2.80 per million output tokens, undercutting both Claude and GPT-4 on a per-token basis. The model is also topping several coding benchmarks, outperforming the big names on tasks developers actually care about.

The Hacker News thread has been buzzing for two days, racking up 870+ points. That’s rare staying power for a model launch. What matters here is not just another model release. It’s the speed at which capability is getting cheaper. Six months ago, this level of performance at this price would have been unthinkable. Now it’s table stakes.

For anyone running a business, this is the cost curve you need to track. The models you’re evaluating today will be half the price, or twice as good, in a quarter. That makes vendor lock-in dangerous and waiting for perfection expensive. The smarter play is to build systems that can swap models in and out as the market moves, which is exactly the kind of flexibility we design into an AI command centre. You want infrastructure that treats models as commodities, not dependencies.

The open-weight angle also matters. If you have the technical chops or the right partner, you can run Kimi K3 on your own hardware, which changes the economics again. You’re no longer paying per token. You’re paying for compute time, and you control the data flow. That’s a different risk profile, especially for companies handling sensitive information or high query volumes.

The takeaway: AI capability is not scarce anymore. Distribution and integration are. If you’re still shopping for the perfect model, you’re solving yesterday’s problem.