We shipped a frontier model on infrastructure
that couldn't hold the pack. By choice.
The product we shipped was beta, and it carried a warning on the chat from day one: unstable until we secure more GPUs. As soon as we crossed ~100 simultaneous sessions, the errors started — GPU VRAM exhaustion, dropped sessions, the works. If you DM'ed or emailed us, you got the same answer every time: the product is not mature enough to handle your harness — we need to secure more GPUs.
That trade-off was deliberate. We run a frontier open-weights model, ablated and fine-tuned for cyber operations — a multi-trillion-parameter model more capable than anything you can run locally today. That capability comes with a cost of operation we cannot sustain at today's scale. So rather than keep charging for an unstable product, we are stopping v1, building v2 properly, and letting the community in through this waitlist.
Thank you — genuinely. This community is amazing, and building in front of you has been the most fun part of the whole run.
