Kimi K3 is live: Moonshot's 2.8T-parameter model graduates from benchmark hype to production

After weeks of benchmark videos, Kimi K3 is generally available. Now the real test begins.

In plain terms: Moonshot AI's Kimi K3 - a 2.8 trillion-parameter MoE model - is now generally available. The question shifts from "can it benchmark?" to "does it hold up in production?"

Kimi K3 reached GA after a preview period that generated significant social media coverage around benchmark comparisons. The 2.8T parameter count uses sparse MoE activation, so effective compute per token is far lower than the headline number suggests - similar architectural pattern to Mixtral and Qwen 3.8 Max. Moonshot AI had already targeted a $50B pre-IPO valuation with K3 commercial traction as the primary justification.

What matters now isn't the benchmarks from controlled preview demos - it's throughput under load, latency at scale, and real-world reasoning quality against Claude and GPT-4o on production workloads. The enterprise adoption curve over the next 90 days will be the actual signal.

So what: Kimi K3 going live is a market event as much as a model event. For developers: test it against frontier models on your specific domain before assuming benchmark parity translates. For market watchers: K3 adoption rate over the next quarter determines whether Moonshot's IPO story holds or needs a revision.