Model card
Core capabilities, limits, and intended use.
Models
Trained by Sophontic to outperform models up to 60× its size on reasoning evaluations. Our first prototype demonstrates that reasoning need not scale with size, and releases soon alongside the perturbation-paradigm eval kit used to verify the result.
Releasing soon
Release standard
The first prototype is paired with its model card, evaluation report, method notes, and access details. The signal should be inspectable without exposing the training recipe.
At release
When the prototype ships, this page will carry the model card, results on the perturbation-paradigm eval kit, methodology notes, and access details.
Core capabilities, limits, and intended use.
Reasoning results against the perturbation-paradigm rails.
Enough detail to inspect the signal without leaking the training recipe.
Prototype availability and repository links at release.
Contact
For research, capital, and deployment conversations at the frontier of reason.