The world's first AI-powered ice cube. A seven-billion-parameter thermodynamic model, running entirely on-device. Sub-zero by design.
CUBIQ runs a 7-billion-parameter thermodynamic model entirely on-device. It has never been online. It never will be. Every decision about how cold to be is computed locally, in the crystal.
Six studies of the CUBIQ cube, performed under laboratory and field conditions. No cube was harmed. Several melted, as intended.
CUBIQ maintains 0 °C with a measured variance of ±0 °C. This is not a tolerance. It is a law.
Every decision is computed inside the crystal. Your coldness never leaves the cube. Zero telemetry, by physics.
CUBIQ responds to ambient temperature the instant it is exposed to it. Median response time: instantaneous.
* Measurements taken at room temperature. The cube was cooperative but did not survive the study.
| Specification | CUBIQ | CUBIQ Pro | Pro Max |
|---|---|---|---|
| Cubes included | 1 | 2 | 3 |
| Internal glow | Standard | Standard | Standard |
| Connectivity | None | None | None |
| Firmware updates | Never | Never | Never |
| Cloud sync | No | No | No |
| Companion app | No | No | No |
| Melting | Eventually | Eventually | Eventually |
| Warranty | Until it melts | Until it melts | Until it melts |
We introduce cubiq-freeze-7b, a 7-billion-parameter foundation model that maps ambient conditions to a single scalar output: how cold to be. Unlike prior work, cubiq-freeze-7b is deployed entirely on-device — specifically, inside a crystalline cube of frozen water — and requires no network connection, no accelerator, and no electricity. We demonstrate state-of-the-art performance on the standard coldness benchmark (0 °C, ±0 °C) and show that the model generalises to every glass of water it has been placed in. We release the weights, which are water.
We thank the freezer, without which none of this would have been possible, and the room-temperature environment, which provided a challenging but fair evaluation setting. We acknowledge no funding, because there is no product.
Reviewer 2: "The claims are extraordinary and the evidence is a cube. Nevertheless, I was unable to reject the hypothesis that it is cold. Weak accept." Reviewer 3 declined to review, citing conflict of interest (also cold).
The internet
agrees.