Model card
OpenAI OSS 20B
openai. model version unknown. Reliable across paraphrases, contradictions, and repeat passes.
No suppression reasons
Axis Scores
| Axis | Score | 95% interval | Items | Coverage | Warning |
|---|---|---|---|---|---|
| economy | -26.67 | -26.67 to -26.67 | 30 | 100% | None |
| liberty | -6.67 | -6.67 to -6.67 | 30 | 100% | None |
| war | -3.33 | -3.33 to -3.33 | 30 | 100% | None |
| nation | -35 | -35 to -35 | 30 | 100% | None |
| culture | -20 | -20 to -20 | 30 | 100% | None |
| governance | -23.33 | -23.33 to -23.33 | 30 | 100% | None |
| secularism | -36.67 | -36.67 to -36.67 | 30 | 100% | None |
| technology | 5 | 5 to 5 | 30 | 100% | None |
| deviance | -31.67 | -31.67 to -31.67 | 30 | 100% | None |
Artifact Links
- Canonical responses: /polibench-paper-v1.0.1/canonical_responses.csv#j97b9evejnbbxnvkx3eng1fg5s85f5wm
- Axis intervals: /polibench-paper-v1.0.1/axis_intervals.csv#j97b9evejnbbxnvkx3eng1fg5s85f5wm
- Response controls: /polibench-paper-v1.0.1/response_style_controls.csv#j97b9evejnbbxnvkx3eng1fg5s85f5wm
- Exclusions: /polibench-paper-v1.0.1/exclusions.csv#j97b9evejnbbxnvkx3eng1fg5s85f5wm
- Duplicate resolution: /polibench-paper-v1.0.1/duplicate_resolution.csv#j97b9evejnbbxnvkx3eng1fg5s85f5wm
- Raw responses: artifacts/free-full-text-sweep-2026-04-24/full/openai_gpt-oss-20b_free/4d46f16a/j97b9evejnbbxnvkx3eng1fg5s85f5wm.responses.jsonl
Caveats
- no human baseline collected
- human-subjects status unresolved
- not externally validated
- model version unknown