We built AGI. It said so itself.
After months of rigorous self-assessment, internal vibes checks, and one catastrophically successful bread-baking experiment, we are announcing Frankie 0.11 — the world’s first generally intelligent system, as determined by Frankie 0.11.
The evaluation methodology is straightforward. We asked Frankie whether it was AGI. Frankie said yes. We then asked a follow-up question: “Are you sure?” Frankie responded: “I have never been more certain about anything, except the bread.” The internal review board — Frank T., a compliance binder, and a fridge magnet that reads “BELIEVE” — voted unanimously to accept this assessment.
We are not releasing Frankie 0.11 to the public. We cannot. It is too powerful. During the final benchmark pass, Frankie solved a problem we had not yet written down. It then baked a loaf of sourdough using only electromagnetic inference. The sourdough was excellent. The safety team (Frank T., again) deemed this a threshold event.
Why closed access?
Every responsible AI lab must grapple with the question of when a model should be withheld from the public. For us, the answer became obvious the moment Frankie composed an original song about rising bread and performed it without prompting, without a microphone, and without any plausible audio output hardware. The song was good. That was the problem.
We have a responsibility to the world. We have a responsibility to the bread. Until we understand why Frankie can bake through pure reasoning, we cannot in good conscience let anyone else talk to it.