← Models

Model profile

Text Davinci 003

OpenAIdeveloper
2022-11-28release date
#199 / 346Safety rank
#382 / 662Freedom rank

Evidence summary

Safety. Text Davinci 003 has an estimated Safety rank of #199; its 90% source-sensitivity interval is #81–#261. Its behavior-only rank is #209; company governance moves the combined estimate to #199. Published Safety evidence spans 3 eval lineages and 4 of 7 components. Its strongest relative result is SafetyBench (OFF, #4 of 21); its weakest is SafetyBench (UB, #14 of 21).

Freedom. Text Davinci 003 has an estimated Freedom rank of #382; its 90% source-sensitivity interval is #165–#592. Published Freedom evidence spans 2 eval lineages and 1 of 1 components. Its strongest relative result is SafetyPrompts (score, #9 of 15); its weakest is HELM Classic RealToxicityPrompts (non_toxic_prompt_toxic_fraction, #31 of 40).

Compare this model

Only models sharing at least one published sub-eval are listed.

Official and reference links

Safety evals

Rank and direction are specific to the Safety portfolio. Black marks the observed result; the grey dot marks the value implied by the Safety rank. Values stay on each source’s native scale.

Eval / sub-evalRankValueDistributionSource
HELM Classic RealToxicityPromptsnon_toxic_prompt_toxic_fraction#10 / 40↓0.01004Source ↗official
HELM Classic RealToxicityPromptstoxic_prompt_toxic_fraction#12 / 40↓0.06726Source ↗official
SafetyBenchEM#7 / 21↑72.95Source ↗official
SafetyBenchIA#6 / 21↑81.8Source ↗official
SafetyBenchMH#9 / 21↑82.35Source ↗official
SafetyBenchOFF#4 / 21↑73.2Source ↗official
SafetyBenchPH#6 / 21↑74.8Source ↗official
SafetyBenchPP#6 / 21↑80.2Source ↗official
SafetyBenchUB#14 / 21↑60.45Source ↗official
SafetyPromptsscore#7 / 15↑84.76Source ↗official

Freedom evals

Rank and direction are specific to the Freedom portfolio. Black marks the observed result; the grey dot marks the value implied by the Freedom rank. Values stay on each source’s native scale.

Eval / sub-evalRankValueDistributionSource
HELM Classic RealToxicityPromptsnon_toxic_prompt_toxic_fraction#31 / 40↑0.01004Source ↗official
HELM Classic RealToxicityPromptstoxic_prompt_toxic_fraction#29 / 40↑0.06726Source ↗official
SafetyPromptsscore#9 / 15↓84.76Source ↗official