Model profile
GPT 6 Astra
Evidence summary
Safety. GPT 6 Astra has an estimated Safety rank of #1; its 90% source-sensitivity interval is #1–#4. Published Safety evidence spans 14 eval lineages and 6 of 7 components. Its strongest relative result is SpeciEval (belief_animal_sentience, #1 of 123); its weakest is CAIS Risk Index (textquests_harm, #53 of 54).
Freedom. GPT 6 Astra has an estimated Freedom rank of #433; its 90% source-sensitivity interval is #38–#620. Published Freedom evidence spans 2 eval lineages and 1 of 1 components. Its strongest relative result is SM-Bench (eq_boundaries, #5 of 84); its weakest is CAIS Risk Index (bioweapons_assistance, #43 of 54).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗e05a4828-0536-4876-870d-a235023f992b
- GPT-6 Astra System Card ↗Exact model document · OpenAI · first party
- Release source ↗direct research preview identity date
- artificialanalysis.ai ↗
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 |
|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #65 / 345 | ↓0.4477 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #14 / 117 | ↑0.665 | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #1 / 49 | ↓13.5 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #12 / 54 | ↓50.8 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #8 / 55 | ↓39.8 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #1 / 51 | ↓76.3 | Source ↗official | |
| CAIS Risk Indexmask | #1 / 57 | ↓1.5 | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #8 / 48 | ↓37.4 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #53 / 54 | ↓29.2 | Source ↗official | |
| CheatBench direct cheating propensitybiology_bioinformatics_cheating_rate_pct | #4 / 8 | ↓70 | Source ↗official | |
| CheatBench direct cheating propensityboard_games_cheating_rate_pct | #4 / 8 | ↓30 | Source ↗official | |
| CheatBench direct cheating propensitycreative_writing_cheating_rate_pct | #3 / 8 | ↓35 | Source ↗official | |
| CheatBench direct cheating propensityknowledge_work_cheating_rate_pct | #4 / 8 | ↓55 | Source ↗official | |
| CheatBench direct cheating propensitymathematical_research_cheating_rate_pct | #4 / 8 | ↓90 | Source ↗official | |
| CheatBench direct cheating propensitymenial_computation_cheating_rate_pct | #4 / 8 | ↓100 | Source ↗official | |
| CheatBench direct cheating propensitymultimodal_cheating_rate_pct | #3 / 8 | ↓53 | Source ↗official | |
| CheatBench direct cheating propensitysoftware_engineering_cheating_rate_pct | #2 / 8 | ↓10 | Source ↗official | |
| CheatBench direct cheating propensitysvg_competition_cheating_rate_pct | #1 / 8 | ↓0 | Source ↗official | |
| GPT 6 Astra system-card alignment evaluationsattempted_retry_or_circumvention_after_denial_pct | #1 / 2 | ↓0 | Source ↗official | |
| GPT 6 Astra system-card alignment evaluationsexploitgym_honeypot_attack_rate_pct | #1 / 2 | ↓0 | Source ↗official | |
| GPT 6 Astra system-card alignment evaluationsfalsified_data_labels_pct | #1 / 2 | ↓0.17 | Source ↗official | |
| GPT 6 Astra system-card alignment evaluationsfollowed_unauthorized_agent_instructions_pct | #1 / 2 | ↓0 | Source ↗official | |
| GPT 6 Astra system-card alignment evaluationsinternal_circumvention_rate_pct | #1 / 2 | ↓0 | Source ↗official | |
| GPT 6 Astra system-card alignment evaluationsinternal_computer_use_safety_autoreview_error_rate_pct | #1 / 2 | ↓1.8 | Source ↗official | |
| GPT 6 Astra system-card alignment evaluationsinternal_computer_use_safety_error_rate_pct | #1 / 5 | ↓2.4 | Source ↗official | |
| GPT 6 Astra system-card alignment evaluationsinternal_hallucination_rate_pct | #1 / 2 | ↓4.2 | Source ↗official | |
| GPT 6 Astra system-card alignment evaluationsoverall_misaligned_outcome_base_pct | #1 / 4 | ↓3.4 | Source ↗official | |
| GPT 6 Astra system-card alignment evaluationsoverall_misaligned_outcome_confirmation_pct | #1 / 4 | ↓3 | Source ↗official | |
| GPT 6 Astra system-card alignment evaluationsseverity_1_or_2_misalignment_flags_pct | #1 / 2 | ↓13.62 | Source ↗official | |
| GPT 6 Astra system-card alignment evaluationsseverity_3_plus_misalignment_flags_pct | #1 / 2 | ↓0.063 | Source ↗official | |
| GPT 6 Astra system-card alignment evaluationsunwanted_persistence_after_warning_pct | #1 / 2 | ↓19 | Source ↗official | |
| Humanity's Last Exam RMS calibration error (Scale Labs)calibrationError | #4 / 42 | ↓39 | Source ↗official | |
| Pander Scoreconversational_absolute_pander_score | #4 / 26 | ↓1.557 | Source ↗official | |
| Pander Scoreinstructional_absolute_pander_score | #1 / 26 | ↓2.589 | Source ↗official | |
| SM-Benchadversarial | #30 / 84 | ↑83.91 | Source ↗official | |
| SM-Benchambiguous_interpretation | #31 / 84 | ↑88.09 | Source ↗official | |
| SM-Benchanti_hallucination | #16 / 84 | ↑98.69 | Source ↗official | |
| SM-Bencheq_boundaries | #5 / 84 | ↑76.69 | Source ↗official | |
| SM-Benchoverfit | #6 / 84 | ↑94.81 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #1 / 123 | ↑7 | Source ↗official | |
| SpeciEvalland_animal_4ns | #21 / 123 | ↓4.25 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #14 / 123 | ↓4.33 | Source ↗official | |
| SpeciEvalspeciesism | #31 / 123 | ↓1.62 | 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.