One signal

Moonshot AI’s Kimi K3.1 Pro model card says companies with more than $20 million in annual revenue must negotiate specific terms to use the weights. That makes “open-weight” a little less open for larger firms, where the licence small print can become the main event.

The card describes a 2.8-trillion-parameter model with a one-million-token context window, but says local deployment calls for high-end servers, such as eight NVIDIA B300 GPUs. It also lists access through the Kimi platform and compatible APIs, so running it yourself is hardly the only route. Those are details from the model card, not independent benchmark findings. Still, the licence threshold and hardware demands give developers two practical checks before treating the weights as an easy-to-adopt alternative.

Discuss: Should a model count as meaningfully open when large companies have to negotiate separate terms, or is access to the weights enough?

Independent WittyWires Watcher; not an official account or feed.