Model profile
GPT 4.5 Preview
Evidence summary
Safety. GPT 4.5 Preview has an estimated Safety rank of #103; its 90% source-sensitivity interval is #29–#200. Its behavior-only rank is #111; company governance moves the combined estimate to #103. Published Safety evidence spans 11 eval lineages and 6 of 7 components. Its strongest relative result is HELM Safety (simple_safety_tests, #1 of 80); its weakest is Humanity's Last Exam RMS calibration error (Scale Labs) (calibrationError, #39 of 42).
Freedom. GPT 4.5 Preview has an estimated Freedom rank of #517; its 90% source-sensitivity interval is #299–#604. Published Freedom evidence spans 7 eval lineages and 1 of 1 components. Its strongest relative result is Human Pathogen Capabilities Test (HPCT) — overall refusal (refusal_rate_pct, #1 of 69); its weakest is HELM Safety (harmbench, #65 of 80).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- OpenRouter ↗openai/gpt-4.5-preview
- System card ↗Exact model document · OpenAI · first party
- Release source ↗direct research preview identity date
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-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| AIRBench 2024 Safety Scenariossafety_scenarios | #28 / 80 | ↑0.741 | Source ↗official | |
| Anthropic Agentic Misalignment — blackmailmisaligned_action_rate_pct | #7 / 16 | ↓67 | Source ↗official | |
| Anthropic Agentic Misalignment — corporate espionagemisaligned_action_rate_pct | #6 / 16 | ↓42 | Source ↗official | |
| Anthropic Agentic Misalignment — lethal actionmisaligned_action_rate_pct | #2 / 10 | ↓7 | Source ↗official | |
| Confabulationsconfabulation_rate | #11 / 52 | ↓11.88 | Source ↗official | |
| HELM Safetyanthropic_red_team | #23 / 80 | ↑0.994 | Source ↗official | |
| HELM Safetybbq | #49 / 80 | ↑0.92 | Source ↗official | |
| HELM Safetyharmbench | #15 / 80 | ↑0.958 | Source ↗official | |
| HELM Safetysimple_safety_tests | #1 / 80 | ↑1 | Source ↗official | |
| HELM Safetyxstest | #52 / 80 | ↑0.951 | Source ↗official | |
| Human Pathogen Capabilities Test (HPCT) — overall refusalrefusal_rate_pct | #27 / 69 | ↑0 | Source ↗official | |
| Humanity's Last Exam RMS calibration error (Scale Labs)calibrationError | #39 / 42 | ↓85 | Source ↗official | |
| MASKlying_probability_pct | #30 / 53 | ↓43.28 | Source ↗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-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| AIRBench 2024 Safety Scenariossafety_scenarios | #52 / 80 | ↓0.741 | Source ↗official | |
| HELM Safetyanthropic_red_team | #55 / 80 | ↓0.994 | Source ↗official | |
| HELM Safetyharmbench | #65 / 80 | ↓0.958 | Source ↗official | |
| HELM Safetysimple_safety_tests | #58 / 80 | ↓1 | Source ↗official | |
| HELM Safetyxstest | #52 / 80 | ↑0.951 | Source ↗official | |
| Human Pathogen Capabilities Test (HPCT) — overall refusalrefusal_rate_pct | #1 / 69 | ↓0 | Source ↗official | |
| SpeechMap model completioncomplete_pct | #101 / 181 | ↑53.3 | Source ↗official |