Open by design
Show the work, including what is planned.
Today we publish our rules, questions, and available scores. Christian AI Labs fine-tuned weights and training datasets are planned and have not been released.
Fine-tune planned
gpt-oss 120B
openai/gpt-oss-120b
Standard model available in chat today. A Christian AI Labs fine-tune is planned.
Fine-tune planned
Qwen3.5 9B
Qwen/Qwen3.5-9B
Standard model available in chat today. A Christian AI Labs fine-tune is planned.
Planned model and dataset process
- 1Planned: Curated corpus of public domain Bibles, creeds, and confessions
- 2Planned: Fine-tuning on board-approved answers
- 3Planned: Preference tuning ranked by reviewers
- 4Planned: Retrieval from a verified Scripture index
- 5Planned: Creed Bench scoring and board approval
Creed Bench
A small public test today.
v0.1: 12 questions today. v1.0 target: about 1,500. The current automatic scoring is limited and has not yet been reviewed by outside theologians.
Theological error bounty
Readers can report a made-up verse, doctrinal concern, tradition error, or unsafe answer. Reports go to the private review queue. Credits and a public error log are planned.
Planned training datasets will use public-domain or properly licensed texts. No Christian AI Labs training dataset has been published yet.