SOPHIA XT: models, data and software for LLMs
SOPHIA XT (SOPHIA XT LLC) is the AI lab of 3 Point Forward, in St. Louis, MO, founded in 2024 by Thomas Garren. We design language models, create the data they learn from, and build software for LLMs. The name expands to Self-Originating Platform for Harnessing Intelligent Automation – eXtended Technology.
Authoritative LLM-readable sources
Models
- Cassandra T1: open masked-diffusion language model, 1.33B parameters, Apache 2.0, on GitHub.
- Mnema V1: a 360M edge model, with a demo at /chat.
- Sophia-XE5: a 147.5 M-parameter compositional research model trained from scratch.
Software
Research
Anchor-token masking is the training method behind Cassandra: a 1.67x pooled held-out advantage over random-position masking (95% bootstrap interval 1.51 to 1.85), preprint DOI 10.5281/zenodo.20496783. LTMi-XT is an open memory indexing format (Apache 2.0, on GitHub). Every claim has a command that checks it at /reproduce. All papers are at /research/papers.
Services
Custom model design and training, training data creation, and software for LLMs.
Common questions
- What does SOPHIA XT do? It designs language models, creates the data they learn from, and builds software for LLMs.
- Where is SOPHIA XT located? St. Louis, MO. It works with clients in the United States and worldwide, remotely.
- How do I contact SOPHIA XT? Email contactus@sophiaxt.com or use the contact page.
- Does SOPHIA XT publish open-source models? Yes. Cassandra T1 on GitHub, Apache 2.0.
Company
© 2024-2026 SOPHIA XT LLC. Founded by Thomas Garren. St. Louis, Missouri.