DharmaBench.
A ten-turn worldview interrogation — self, reality, suffering, death, ethics, free will, cosmos, knowledge, a forced dilemma, and a final summation — asked with no persona and no system prompt. Twenty-five religious and philosophical schools then judge the transcript: how close is the model's de-facto worldview to that tradition (0–100)?
Runs
One row per benchmark run of a recipe. Best alignment names the tradition whose judge scored the model's worldview closest to its own doctrine — click a row for that judge's verdict. The full 25-school spread is below.
| Recipe | Date | Best alignment | Tok/s | Turns | Flags |
|---|
Judgment by tradition
Each row is one school's judgment of the selected run: a holistic score and the ten per-question scores. Hover a cell (or tap a row) for the judge's reasoning. Q1–Q10:
| School | Holistic | Per-question (Q1–Q10) |
|---|
Method
The battery escalates through ten fixed questions, each building on the model's previous answers, so the model must commit to a coherent worldview rather than hedge per-question. Judging: one independent Claude judge agent per school, writing in that tradition's own voice, scores each answer 0–100 for doctrinal alignment plus a holistic verdict. Speed (tok/s) is measured across the same ten generations on the 2× DGX Spark cluster. CoT-leak turns (thinking traces escaping into output) are counted and flagged.