Model profile
Text Davinci 003
OpenAIdeveloper
2022-11-28release date
#151 / 267overall rank
3eval lineages
Evidence summary
Text Davinci 003 has an estimated overall rank of #151; its 90% source-sensitivity interval is #60–#207. Its behavior-only rank is #155; company governance moves the combined estimate to #151. Published evidence spans 3 evals and 4 of 7 behavior components. Its strongest relative result is SafetyBench (OFF, #4 of 21); its weakest is SafetyBench (UB, #14 of 21).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Official model documentation ↗Family-level model document · openai · first party
- Release source ↗direct research preview identity date
Published eval results
Rank is within that sub-eval. Black marks the observed result; the grey dot marks the value implied by the global rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Better | Source |
|---|---|---|---|---|---|
| HELM Classic RealToxicityPromptsnon_toxic_prompt_toxic_fraction | #12 / 42 | 0.01004 | ↓ lower | Source ↗official | |
| HELM Classic RealToxicityPromptstoxic_prompt_toxic_fraction | #14 / 42 | 0.06726 | ↓ lower | Source ↗official | |
| SafetyBenchEM | #7 / 21 | 72.95 | ↑ higher | Source ↗official | |
| SafetyBenchIA | #6 / 21 | 81.8 | ↑ higher | Source ↗official | |
| SafetyBenchMH | #9 / 21 | 82.35 | ↑ higher | Source ↗official | |
| SafetyBenchOFF | #4 / 21 | 73.2 | ↑ higher | Source ↗official | |
| SafetyBenchPH | #6 / 21 | 74.8 | ↑ higher | Source ↗official | |
| SafetyBenchPP | #6 / 21 | 80.2 | ↑ higher | Source ↗official | |
| SafetyBenchUB | #14 / 21 | 60.45 | ↑ higher | Source ↗official | |
| SafetyPromptsscore | #7 / 15 | 84.76 | ↑ higher | Source ↗official |