Tier 3 Practical open-weight
What the publisher claims: open-source, reconstruction
The one-liner: The community's answer to a closed decision model: the idea, rebuilt, minus the intelligence.
The six questions
Code and (random) weights on GitHub, ungated.
Apache 2.0.
Standard PyTorch on user hardware.
Reconstructs the published idea (typed outputs, parallel sampling); no training data or method.
By its own README it does not reproduce the production model — the idea is open, the system is not.
Community project accepting outside contributions, though single-maintainer.
License
Apache 2.0 OSI-approved — license source ↗
- Random weights — explicitly not the production Jev model
- Makes no claim of matching TypeSafe's published results
What is actually released
- PyTorch implementation of the JEVA-style architecture
- example code
- random-weight checkpoints
Why it matters
The paired control to Jev: when a category-defining model stays closed, the open ecosystem reconstructs the interface — and the reconstruction is honest that it is not equivalent. Interface compatibility is not model equivalence.
Sources
Derived from the openness dataset behind But Is It Open? — snapshot 2026-11-07 ↗. Scores are qualitative readings of the cited evidence, not verdicts; tier measures openness, not intelligence.