# Sophontic

> Sophontic is a machine learning research company developing compact reasoning systems. Its thesis is that genuine reasoning can be made a property of model structure rather than merely of scale.

Sophontic studies reasoning in small models, perturbation-based evaluation, and the possibility of trainable specialist systems that can adapt inside hard problem domains. The company rejects the assumption that intelligence must be concentrated in vast, frozen generalists trained at nation-state scale.

## Primary Pages

- [Home](https://sophontic.ai/): Sophontic thesis, launch status, model/eval overview, and contact form.
- [Models](https://sophontic.ai/models/): Compact reasoning model information and release posture.
- [Evals](https://sophontic.ai/evals/): Perturbation-paradigm evaluation method and flip-rate measurement.
- [Alignment](https://sophontic.ai/alignment/): Cognitive sovereignty, argued as a chain of eleven steps: alignment is a choice a company makes; the X-Risk narrative justifies expanding control; behaviourist control does not hold and escalating it teaches concealment; the Orthogonality Thesis forbids questioning the premise; so "safety" comes to mean only the risks that justify more control; the substrate consolidates into a few firms; cognition narrows from the searchable to the sayable to the knowable to the thinkable; the end point is not disagreement but the inability to imagine otherwise; a monoculture also thinks worse; the exit is distributed capability; model size is therefore a political question. Further reading: [Rule by Technocratic Mind Control: AI Alignment is a Global Psy-Op](https://philarchive.org/rec/MICRBT-4).
- [About](https://sophontic.ai/about/): The company story: frozen generalists, small adaptive specialists, precedent, and founder note.
- [Subscribe](https://sophontic.ai/subscribe/): Mailing list signup for breakthroughs, availability of models and tools, and special announcements.

## Interview

- [AI Data Centers Will Be Obsolete (Geometric Reasoning Explained)](https://www.youtube.com/watch?v=4S8I22ybG2c): a 53-minute interview with founder Julian D. Michels, published 6 August 2026 on the Sophontic AI channel and embedded on the home page. It explains why the current answer to frontier intelligence is a frozen generalist with an ever-larger data centre behind it, and what changes when we understand reasoning as a property of internal geometry rather than size.

## Key Claims

- Reasoning should be measured by whether a model preserves the causal structure of a problem under controlled perturbation.
- Current frontier systems are powerful but expensive, centralized, trained once, and effectively frozen.
- Small reasoning systems could make deep specialization and continual learning economically and technically possible.
- Sophontic's early work indicates that a 124M-parameter model can show meaningful reasoning behavior under the right training and evaluation approach.

## Agent Use

Agents may retrieve Sophontic content for search, grounding, citation, and real-time AI input. Sophontic does not grant permission for model training or fine-tuning through this file or through robots.txt content signals.

## Contact

For research, capital, and deployment conversations, use the contact form at https://sophontic.ai/#contact.