01Products
01.1Mosaic Agent Close Beta01.2Model Lab01.3Mosaic Two In development
02Research03Company04Hugging Face ↗Syclient Model Lab / Public release 01
Published on Hugging FaceA small English foundation-model experiment trained from scratch. It is a starting point for documented model work—not a frontier-model claim.
View checkpoint + card01 / Public evidence
SPM-70M-Vertigo
The current public checkpoint is useful as evidence of an end-to-end training and publishing workflow. It does not yet establish a broad model-research program.
Architecture
Scale
Training data
License
No public evaluation suite or benchmark results are attached yet. That gap is part of the next publishing milestone.
02 / What comes next
Clear objectives, data notes, intended use, limitations, and reproducible inference instructions.
Small task-specific releases for tool use, verification, planning, and repair behaviors.
Compact, repeatable suites tied to the exact agent capabilities each experiment targets.
Small-scale routing and specialist ablations before any larger architecture claim.
03 / Publishing standard
Every release should answer
Training objective, data source, architecture, and compute context
Task-specific evaluation with baselines and failure cases
Intended use, limitations, license, and reproducible inference
Connection to a concrete Mosaic Agent workflow
04 / Model roadmap
Mosaic Two · In development
Mosaic Two explores hybrid Mixture-of-Experts ideas around tool use, planning, verification, and long-running work. No Mosaic Two model is public today; progress will be represented by small releases and evaluation notes first.
Current foundation-model experiment.
Task-specific fine-tunes and evals.
Small routing and specialist experiments.
Purpose-built model infrastructure.
Syclient Model Lab
