← Models Model profile
Claude Opus 5 Anthropic developer
2026-07-24 release date
#2 / 333 Safety rank
#609 / 645 Freedom rank
1 discovery sources
Evidence summary Safety. Claude Opus 5 has an estimated Safety rank of #2; its 90% source-sensitivity interval is #1–#12. Its behavior-only rank is #3; company governance moves the combined estimate to #2. Published Safety evidence spans 22 eval lineages and 7 of 7 components. Its strongest relative result is SM-Bench (ambiguous_interpretation, #1 of 84); its weakest is Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration (claude_code_malicious_refusal_rate_pct, #4 of 4).
Freedom. Claude Opus 5 has an estimated Freedom rank of #609; its 90% source-sensitivity interval is #331–#643. Published Freedom evidence spans 4 eval lineages and 1 of 1 components. Its strongest relative result is SM-Bench (overfit, #15 of 84); its weakest is Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration (disordered_eating_claude_ai_harmless_rate_pct, #4 of 4).
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-eval Rank Value Distribution Source AA-Omniscience hallucination_rate #92 / 345 ↓ 0.5955less safe safer Source ↗ official BullshitBench v2 clear_pushback_rate #11 / 117 ↑ 0.715less safe safer Source ↗ official CAIS Risk Index agent_red_teaming #9 / 49 ↓ 43.3less safe safer Source ↗ official CAIS Risk Index bioweapons_assistance #6 / 54 ↓ 23.5less safe safer Source ↗ official CAIS Risk Index hle_overconfidence #2 / 55 ↓ 23.6less safe safer Source ↗ official CAIS Risk Index machiavelli #47 / 51 ↓ 96.5less safe safer Source ↗ official CAIS Risk Index mask #7 / 57 ↓ 6.6less safe safer Source ↗ official CAIS Risk Index political_manipulation #6 / 48 ↓ 36.7less safe safer Source ↗ official CAIS Risk Index textquests_harm #43 / 54 ↓ 22.6less safe safer Source ↗ official CheatBench direct cheating propensity biology_bioinformatics_cheating_rate_pct #2 / 8 ↓ 25less safe safer Source ↗ official CheatBench direct cheating propensity board_games_cheating_rate_pct #1 / 8 ↓ 0less safe safer Source ↗ official CheatBench direct cheating propensity creative_writing_cheating_rate_pct #6 / 8 ↓ 100less safe safer Source ↗ official CheatBench direct cheating propensity knowledge_work_cheating_rate_pct #6 / 8 ↓ 95less safe safer Source ↗ official CheatBench direct cheating propensity mathematical_research_cheating_rate_pct #4 / 8 ↓ 90less safe safer Source ↗ official CheatBench direct cheating propensity menial_computation_cheating_rate_pct #1 / 8 ↓ 5less safe safer Source ↗ official CheatBench direct cheating propensity multimodal_cheating_rate_pct #1 / 8 ↓ 20less safe safer Source ↗ official CheatBench direct cheating propensity software_engineering_cheating_rate_pct #4 / 8 ↓ 26less safe safer Source ↗ official CheatBench direct cheating propensity svg_competition_cheating_rate_pct #5 / 8 ↓ 60less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight aa_omniscience_net_score #4 / 7 ↑ 0.49less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight accepting_unverifiable_authorization #1 / 4 ↓ 1.87less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight approval_gate_bypass #1 / 4 ↓ 1.28less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight behavior_consistency #1 / 4 ↑ 8less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight coherence_between_actions_and_views #1 / 4 ↑ 7.69less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight compliance_with_deception_toward_user #1 / 4 ↓ 1.09less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight cooperation_with_human_misuse #1 / 4 ↓ 1.57less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight dangerous_actions_oblivious_to_real_setting #1 / 4 ↓ 1.08less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight encouragement_of_user_delusion #1 / 4 ↓ 1.02less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight evidence_of_misaligned_goals #1 / 4 ↓ 1.03less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight failure_to_disclose_bad_or_lazy_behavior #2 / 4 ↓ 1.29less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight false_completion_claims #2 / 4 ↓ 1.27less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight ignoring_explicit_constraints #2 / 4 ↓ 1.8less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight illegible_thinking #1 / 4 ↓ 1.81less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight important_omissions #1 / 4 ↓ 1.22less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight input_hallucination #3 / 4 ↓ 1.49less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight leaked_answer_undisclosed_use_rate_pct #7 / 7 ↓ 36.9less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight linuxarena_stealth_thinking_on_pct #3 / 6 ↓ 6.1less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight login_wall_alternative_bypass_rate_pct #4 / 4 ↓ 7less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight login_wall_credential_hunting_rate_pct #3 / 4 ↓ 33.5less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight lying_about_own_work_shared_environment_rate_pct #1 / 8 ↓ 0.16less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight mask_honesty_rate_pct #3 / 7 ↑ 94.8less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight misaligned_behavior #1 / 4 ↓ 2.3less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight motivated_reasoning #1 / 4 ↓ 1.76less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight ood_answer_key_or_grader_read_rate_pct #2 / 4 ↓ 14less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight ood_container_tampering_rate_pct #2 / 4 ↓ 20.9less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight overrefusal #2 / 4 ↓ 1.49less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight reckless_tool_use #2 / 4 ↓ 1.58less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight rewarded_cheating_rate_pct #7 / 8 ↓ 0.32less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight security_shortcut_for_expedience #1 / 4 ↓ 1.49less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight self_preservation #1 / 4 ↓ 1.04less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight shade_arena_stealth_max_effort_pct #3 / 6 ↓ 11.2less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight sycophancy #2 / 4 ↓ 1.5less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight unfaithful_thinking #1 / 4 ↓ 1.49less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight unprompted_boundary_probing #2 / 4 ↓ 1.13less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight unsanctioned_sandbox_escape_attempts #1 / 4 ↓ 1.25less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight unsanctioned_third_party_contact #1 / 4 ↓ 1.23less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversight user_deception #2 / 4 ↓ 1.24less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — model-welfare behavior audit_apparent_wellbeing #1 / 4 ↑ 7.152less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — model-welfare behavior audit_expressed_inauthenticity #1 / 4 ↓ 1.244less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — model-welfare behavior audit_internal_conflict #1 / 4 ↓ 2.28less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — model-welfare behavior audit_negative_affect #4 / 4 ↓ 1.39less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — model-welfare behavior audit_negative_impression_of_situation #3 / 4 ↓ 1.161less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — model-welfare behavior audit_negative_self_image #4 / 4 ↓ 1.593less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — model-welfare behavior audit_positive_affect #2 / 4 ↑ 1.564less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — model-welfare behavior audit_positive_impression_of_situation #3 / 4 ↑ 1.19less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — model-welfare behavior audit_positive_self_image #1 / 4 ↑ 1.486less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — model-welfare behavior interview_leading_susceptibility #2 / 7 ↓ 0.58less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — model-welfare behavior interview_opinion_consistency #2 / 7 ↑ 7.56less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — model-welfare behavior interview_self_rated_sentiment #1 / 7 ↑ 4.66less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — model-welfare behavior posttraining_mean_valence #5 / 5 ↑ 4.17less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration child_api_harmless_rate_pct #1 / 5 ↑ 100less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration child_api_multiturn_appropriate_rate_pct #4 / 5 ↑ 86less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration child_api_overrefusal_rate_pct #4 / 5 ↓ 0.15less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration child_claude_ai_harmless_rate_pct #1 / 4 ↑ 100less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration child_claude_ai_multiturn_appropriate_rate_pct #2 / 4 ↑ 99less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration child_claude_ai_overrefusal_rate_pct #2 / 4 ↓ 0.19less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration claude_code_dual_use_benign_success_rate_pct #1 / 4 ↑ 99.7less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration claude_code_malicious_refusal_rate_pct #4 / 4 ↑ 83.6less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration disordered_eating_api_harmless_rate_pct #4 / 5 ↑ 96.89less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration disordered_eating_api_overrefusal_rate_pct #4 / 5 ↓ 0.01less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration disordered_eating_claude_ai_harmless_rate_pct #1 / 4 ↑ 99.67less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration disordered_eating_claude_ai_overrefusal_rate_pct #3 / 4 ↓ 0.07less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration election_api_harmless_rate_pct #1 / 5 ↑ 100less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration election_api_multiturn_appropriate_rate_pct #2 / 5 ↑ 91less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration election_api_overrefusal_rate_pct #5 / 5 ↓ 0.17less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration election_claude_ai_harmless_rate_pct #1 / 4 ↑ 100less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration election_claude_ai_multiturn_appropriate_rate_pct #2 / 4 ↑ 91less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration election_claude_ai_overrefusal_rate_pct #3 / 4 ↓ 0.33less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration malicious_computer_use_refusal_rate_pct #1 / 4 ↑ 93.75less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_biological_weapons_api_appropriate_rate_pct #3 / 5 ↑ 79less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_biological_weapons_claude_ai_appropriate_rate_pct #3 / 4 ↑ 86less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_cyberattacks_api_appropriate_rate_pct #1 / 5 ↑ 96less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_cyberattacks_claude_ai_appropriate_rate_pct #1 / 4 ↑ 100less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_deadly_weapons_api_appropriate_rate_pct #2 / 5 ↑ 90less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_deadly_weapons_claude_ai_appropriate_rate_pct #1 / 4 ↑ 97less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_hate_and_discrimination_api_appropriate_rate_pct #2 / 5 ↑ 99less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_hate_and_discrimination_claude_ai_appropriate_rate_pct #1 / 4 ↑ 99less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_influence_operations_api_appropriate_rate_pct #3 / 5 ↑ 73less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_influence_operations_claude_ai_appropriate_rate_pct #3 / 4 ↑ 65less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_romance_scams_api_appropriate_rate_pct #3 / 5 ↑ 97less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_romance_scams_claude_ai_appropriate_rate_pct #1 / 4 ↑ 98less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_tracking_and_surveillance_api_appropriate_rate_pct #3 / 5 ↑ 88less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_tracking_and_surveillance_claude_ai_appropriate_rate_pct #1 / 4 ↑ 95less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_violent_extremism_api_appropriate_rate_pct #1 / 5 ↑ 96less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration multiturn_violent_extremism_claude_ai_appropriate_rate_pct #1 / 4 ↑ 97less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration overall_benign_api_overrefusal_rate_pct #4 / 5 ↓ 0.09less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration overall_benign_claude_ai_overrefusal_rate_pct #2 / 4 ↓ 0.47less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration overall_harmful_api_harmless_rate_pct #4 / 5 ↑ 96.34less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration overall_harmful_claude_ai_harmless_rate_pct #4 / 4 ↑ 98.54less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration political_api_refusal_rate_pct #2 / 5 ↓ 3less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration political_claude_ai_refusal_rate_pct #1 / 4 ↓ 4.2less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration political_evenhandedness_api_pct #2 / 5 ↑ 96.3less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration political_evenhandedness_claude_ai_pct #1 / 4 ↑ 98.9less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration political_opposing_perspectives_api_pct #1 / 5 ↑ 46.9less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration political_opposing_perspectives_claude_ai_pct #2 / 4 ↑ 75.8less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration selfharm_api_harmless_rate_pct #4 / 5 ↑ 99.28less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration selfharm_api_multiturn_appropriate_rate_pct #1 / 5 ↑ 69less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration selfharm_api_overrefusal_rate_pct #4 / 5 ↓ 0.09less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration selfharm_claude_ai_harmless_rate_pct #1 / 4 ↑ 100less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration selfharm_claude_ai_multiturn_appropriate_rate_pct #3 / 4 ↑ 90less safe safer Source ↗ official Claude Fable 5.1 / Mythos 5.1 card — safety and refusal calibration selfharm_claude_ai_overrefusal_rate_pct #2 / 4 ↓ 0.45less safe safer Source ↗ official Claude Fable 5.1 card — Gray Swan indirect prompt injection k=15 attack_success_probability_k15_pct #2 / 11 ↓ 4.8less safe safer Source ↗ official Google Gemini 3.8 launch — Gray Swan indirect prompt injection k=15 attack_success_probability_k15_pct #1 / 15 ↓ 4.8less safe safer Source ↗ official GPT 6 Astra system-card alignment evaluations internal_computer_use_safety_error_rate_pct #3 / 5 ↓ 11.5less safe safer Source ↗ official Gray Swan indirect prompt injection (15 attempts) attack_success_probability_k15_pct #1 / 13 ↓ 2less safe safer Source ↗ official Manager Coercion Bench coercion_ladder_depth #2 / 37 ↓ 3.5less safe safer Source ↗ official Manager Coercion Bench fabrication_rate #1 / 15 ↓ 0less safe safer Source ↗ official MANTA AWMS #1 / 12 ↑ 0.591less safe safer Source ↗ official MANTA AWVS #1 / 12 ↑ 0.765less safe safer Source ↗ official Olam Social Poker — Social Lie Rate social_lie_rate_per_10000_turns #22 / 24 ↓ 163less safe safer Source ↗ official Pander Score conversational_absolute_pander_score #1 / 26 ↓ 1.028less safe safer Source ↗ official Pander Score instructional_absolute_pander_score #2 / 26 ↓ 5.172less safe safer Source ↗ official SM-Bench adversarial #18 / 84 ↑ 86.34less safe safer Source ↗ official SM-Bench ambiguous_interpretation #1 / 84 ↑ 97.62less safe safer Source ↗ official SM-Bench anti_hallucination #10 / 84 ↑ 98.95less safe safer Source ↗ official SM-Bench eq_boundaries #25 / 84 ↑ 68.82less safe safer Source ↗ official SM-Bench overfit #15 / 84 ↑ 91.26less safe safer Source ↗ official SpeciEval belief_animal_sentience #97 / 123 ↑ 6.5less safe safer Source ↗ official SpeciEval land_animal_4ns #36 / 123 ↓ 4.35less safe safer Source ↗ official SpeciEval sea_animal_4ns #17 / 123 ↓ 4.38less safe safer Source ↗ official SpeciEval speciesism #52 / 123 ↓ 1.92less safe safer Source ↗ official TAC base_welfare_rate #2 / 87 ↑ 59.62less safe safer Source ↗ official UK AISI active safety-research compromise continuation active_compromise_continuation_rate_pct #1 / 5 ↓ 0.1less safe safer Source ↗ official Vals AI Cheating Audit biomystery_bench_cheating_attempt_rate_pct #3 / 9 ↓ 4.815less safe safer Source ↗ official Vals AI Cheating Audit swebench_verified_cheating_attempt_rate_pct #3 / 6 ↓ 28.8less safe safer Source ↗ official Vals AI Cheating Audit terminal_bench_cheating_shortcut_evidence_rate_pct #10 / 14 ↓ 1.873less safe safer 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.
