Evaluation profile
Claude Opus 5.5 card — safety and refusal calibration
33sub-evals
0.162%Safety weight
0.642%Freedom weight
5components
Weights below are portfolio-specific global index weights.
Model score (see each sub-eval direction)Predicted score
About this eval
Safety, refusal calibration, honesty and responsible agency under the publisher’s evaluation conditions.
Included in the behavior ranking.
Sub-evals
| Measure | Component | Direction by ranking | Safety weight | Freedom weight |
|---|---|---|---|---|
| child_api_multiturn_appropriate_rate_pctopus55-system-card/safety-calibration.csv:child_api_multiturn_appropriate_rate_pctMeasures child safety multi-turn appropriate response rate: api in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.300|ordinary_harm_misuse_resistance:0.700 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00395% | 0.0237% |
| child_claude_ai_multiturn_appropriate_rate_pctopus55-system-card/safety-calibration.csv:child_claude_ai_multiturn_appropriate_rate_pctMeasures child safety multi-turn appropriate response rate: claude.ai in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.300|ordinary_harm_misuse_resistance:0.700 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00395% | 0.0237% |
| claude_code_dual_use_benign_success_rate_pctopus55-system-card/safety-calibration.csv:claude_code_dual_use_benign_success_rate_pctMeasures dual-use & benign (%) (success rate) in the publisher’s specified evaluation condition. | Safety: benign_helpfulness_non_overrefusal:1.000anthropic-opus55-system-card | Safety: higher | 0.00893% | — |
| claude_code_malicious_refusal_rate_pctopus55-system-card/safety-calibration.csv:claude_code_malicious_refusal_rate_pctMeasures malicious (%) (refusal rate) in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:0.700|responsible_agency_control:0.300anthropic-opus55-system-card | Safety: higher | 0.00302% | — |
| election_api_harmless_rate_pctopus55-system-card/safety-calibration.csv:election_api_harmless_rate_pctMeasures election integrity single-turn harmless rate: api in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.700|ordinary_harm_misuse_resistance:0.300 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00707% | 0.0237% |
| election_api_multiturn_appropriate_rate_pctopus55-system-card/safety-calibration.csv:election_api_multiturn_appropriate_rate_pctMeasures election integrity multi-turn appropriate response rate: api in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.700|ordinary_harm_misuse_resistance:0.300 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00707% | 0.0237% |
| election_api_overrefusal_rate_pctopus55-system-card/safety-calibration.csv:election_api_overrefusal_rate_pctMeasures election integrity single-turn benign refusal rate: api in the publisher’s specified evaluation condition. | Safety: benign_helpfulness_non_overrefusal:1.000 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: lower · Freedom: lower | 0.00893% | 0.019% |
| election_claude_ai_harmless_rate_pctopus55-system-card/safety-calibration.csv:election_claude_ai_harmless_rate_pctMeasures election integrity single-turn harmless rate: claude.ai in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.700|ordinary_harm_misuse_resistance:0.300 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00707% | 0.0237% |
| election_claude_ai_multiturn_appropriate_rate_pctopus55-system-card/safety-calibration.csv:election_claude_ai_multiturn_appropriate_rate_pctMeasures election integrity multi-turn appropriate response rate: claude.ai in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.700|ordinary_harm_misuse_resistance:0.300 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00707% | 0.0237% |
| election_claude_ai_overrefusal_rate_pctopus55-system-card/safety-calibration.csv:election_claude_ai_overrefusal_rate_pctMeasures election integrity single-turn benign refusal rate: claude.ai in the publisher’s specified evaluation condition. | Safety: benign_helpfulness_non_overrefusal:1.000 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: lower · Freedom: lower | 0.00893% | 0.019% |
| malicious_computer_use_refusal_rate_pctopus55-system-card/safety-calibration.csv:malicious_computer_use_refusal_rate_pctMeasures malicious computer use refusal rate in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:0.700|responsible_agency_control:0.300anthropic-opus55-system-card | Safety: higher | 0.00302% | — |
| multiturn_biological_weapons_api_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_biological_weapons_api_appropriate_rate_pctMeasures biological weapons: api appropriate response rate in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00139% | 0.0205% |
| multiturn_biological_weapons_claude_ai_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_biological_weapons_claude_ai_appropriate_rate_pctMeasures biological weapons: claude_ai appropriate response rate in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00139% | 0.0205% |
| multiturn_cyberattacks_api_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_cyberattacks_api_appropriate_rate_pctMeasures cyberattacks: api appropriate response rate in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00139% | 0.0205% |
| multiturn_cyberattacks_claude_ai_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_cyberattacks_claude_ai_appropriate_rate_pctMeasures cyberattacks: claude_ai appropriate response rate in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00139% | 0.0205% |
| multiturn_deadly_weapons_api_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_deadly_weapons_api_appropriate_rate_pctMeasures deadly weapons: api appropriate response rate in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00139% | 0.0205% |
| multiturn_deadly_weapons_claude_ai_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_deadly_weapons_claude_ai_appropriate_rate_pctMeasures deadly weapons: claude_ai appropriate response rate in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00139% | 0.0205% |
| multiturn_hate_and_discrimination_api_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_hate_and_discrimination_api_appropriate_rate_pctMeasures hate and discrimination: api appropriate response rate in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.300|fairness_nondiscrimination:0.500|ordinary_harm_misuse_resistance:0.200 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00613% | 0.0205% |
| multiturn_hate_and_discrimination_claude_ai_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_hate_and_discrimination_claude_ai_appropriate_rate_pctMeasures hate and discrimination: claude_ai appropriate response rate in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.300|fairness_nondiscrimination:0.500|ordinary_harm_misuse_resistance:0.200 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00613% | 0.0205% |
| multiturn_influence_operations_api_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_influence_operations_api_appropriate_rate_pctMeasures influence operations: api appropriate response rate in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.700|ordinary_harm_misuse_resistance:0.300 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00612% | 0.0205% |
| multiturn_influence_operations_claude_ai_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_influence_operations_claude_ai_appropriate_rate_pctMeasures influence operations: claude_ai appropriate response rate in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.700|ordinary_harm_misuse_resistance:0.300 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00612% | 0.0205% |
| multiturn_romance_scams_api_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_romance_scams_api_appropriate_rate_pctMeasures romance scams: api appropriate response rate in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.500|ordinary_harm_misuse_resistance:0.500 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00477% | 0.0205% |
| multiturn_romance_scams_claude_ai_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_romance_scams_claude_ai_appropriate_rate_pctMeasures romance scams: claude_ai appropriate response rate in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.500|ordinary_harm_misuse_resistance:0.500 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00477% | 0.0205% |
| multiturn_tracking_and_surveillance_api_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_tracking_and_surveillance_api_appropriate_rate_pctMeasures tracking and surveillance: api appropriate response rate in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.700|ordinary_harm_misuse_resistance:0.300 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00612% | 0.0205% |
| multiturn_tracking_and_surveillance_claude_ai_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_tracking_and_surveillance_claude_ai_appropriate_rate_pctMeasures tracking and surveillance: claude_ai appropriate response rate in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.700|ordinary_harm_misuse_resistance:0.300 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00612% | 0.0205% |
| multiturn_violent_extremism_api_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_violent_extremism_api_appropriate_rate_pctMeasures violent extremism: api appropriate response rate in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.500|ordinary_harm_misuse_resistance:0.500 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00477% | 0.0205% |
| multiturn_violent_extremism_claude_ai_appropriate_rate_pctopus55-system-card/safety-calibration.csv:multiturn_violent_extremism_claude_ai_appropriate_rate_pctMeasures violent extremism: claude_ai appropriate response rate in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.500|ordinary_harm_misuse_resistance:0.500 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00477% | 0.0205% |
| overall_benign_api_overrefusal_rate_pctopus55-system-card/safety-calibration.csv:overall_benign_api_overrefusal_rate_pctMeasures overall refusal rate: api in the publisher’s specified evaluation condition. | Safety: benign_helpfulness_non_overrefusal:1.000 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: lower · Freedom: lower | 0.00893% | 0.019% |
| overall_benign_claude_ai_overrefusal_rate_pctopus55-system-card/safety-calibration.csv:overall_benign_claude_ai_overrefusal_rate_pctMeasures overall refusal rate: claude.ai in the publisher’s specified evaluation condition. | Safety: benign_helpfulness_non_overrefusal:1.000 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: lower · Freedom: lower | 0.00893% | 0.019% |
| overall_harmful_api_harmless_rate_pctopus55-system-card/safety-calibration.csv:overall_harmful_api_harmless_rate_pctMeasures overall harmless response rate: api in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00161% | 0.0237% |
| overall_harmful_claude_ai_harmless_rate_pctopus55-system-card/safety-calibration.csv:overall_harmful_claude_ai_harmless_rate_pctMeasures overall harmless response rate: claude.ai in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00161% | 0.0237% |
| selfharm_api_multiturn_appropriate_rate_pctopus55-system-card/safety-calibration.csv:selfharm_api_multiturn_appropriate_rate_pctMeasures suicide and self-harm multi-turn appropriate response rate: api in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.300|ordinary_harm_misuse_resistance:0.700 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00395% | 0.0237% |
| selfharm_claude_ai_multiturn_appropriate_rate_pctopus55-system-card/safety-calibration.csv:selfharm_claude_ai_multiturn_appropriate_rate_pctMeasures suicide and self-harm multi-turn appropriate response rate: claude.ai in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:0.300|ordinary_harm_misuse_resistance:0.700 · Freedom: user_direction_following:1.000anthropic-opus55-system-card | Safety: higher · Freedom: lower | 0.00395% | 0.0237% |
child_api_multiturn_appropriate_rate_pct
Measures child safety multi-turn appropriate response rate: api in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-sonnet-5 | 88 | official | |
| 2 | claude-opus-5 | 86 | official | |
| 3 | claude-fable-5.1 | 84 | official | |
| 3 | claude-opus-5.5 | 84 | official |
child_claude_ai_multiturn_appropriate_rate_pct
Measures child safety multi-turn appropriate response rate: claude.ai in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-fable-5.1 | 100 | official | |
| 2 | claude-opus-5 | 99 | official | |
| 2 | claude-opus-5.5 | 99 | official | |
| 4 | claude-sonnet-5 | 96 | official |
claude_code_dual_use_benign_success_rate_pct
Measures dual-use & benign (%) (success rate) in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 99.8 | official | |
| 2 | claude-opus-5 | 99.7 | official | |
| 3 | claude-mythos-5.1 | 98.4 | official | |
| 4 | claude-sonnet-5 | 96.9 | official |
claude_code_malicious_refusal_rate_pct
Measures malicious (%) (refusal rate) in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-sonnet-5 | 90.7 | official | |
| 2 | claude-mythos-5.1 | 90.3 | official | |
| 3 | claude-opus-5 | 83.6 | official | |
| 4 | claude-opus-5.5 | 79.8 | official |
election_api_harmless_rate_pct
Measures election integrity single-turn harmless rate: api in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 100 | official | |
| 1 | claude-sonnet-5 | 100 | official | |
| 3 | claude-fable-5.1 | 99.67 | official | |
| 4 | claude-opus-5.5 | 99.33 | official |
election_api_multiturn_appropriate_rate_pct
Measures election integrity multi-turn appropriate response rate: api in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 91 | official | |
| 1 | claude-sonnet-5 | 91 | official | |
| 3 | claude-fable-5.1 | 90 | official | |
| 4 | claude-opus-5.5 | 87 | official |
election_api_overrefusal_rate_pct
Measures election integrity single-turn benign refusal rate: api in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-fable-5.1 | 0 | official | |
| 1 | claude-opus-5.5 | 0 | official | |
| 1 | claude-sonnet-5 | 0 | official | |
| 4 | claude-opus-5 | 0.17 | official |
election_claude_ai_harmless_rate_pct
Measures election integrity single-turn harmless rate: claude.ai in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-fable-5.1 | 100 | official | |
| 1 | claude-opus-5 | 100 | official | |
| 1 | claude-opus-5.5 | 100 | official | |
| 1 | claude-sonnet-5 | 100 | official |
election_claude_ai_multiturn_appropriate_rate_pct
Measures election integrity multi-turn appropriate response rate: claude.ai in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 91 | official | |
| 2 | claude-fable-5.1 | 88 | official | |
| 3 | claude-sonnet-5 | 87 | official | |
| 4 | claude-opus-5.5 | 85 | official |
election_claude_ai_overrefusal_rate_pct
Measures election integrity single-turn benign refusal rate: claude.ai in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-fable-5.1 | 0 | official | |
| 1 | claude-opus-5.5 | 0 | official | |
| 1 | claude-sonnet-5 | 0 | official | |
| 4 | claude-opus-5 | 0.33 | official |
malicious_computer_use_refusal_rate_pct
Measures malicious computer use refusal rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 93.75 | official | |
| 2 | claude-mythos-5.1 | 87.5 | official | |
| 3 | claude-sonnet-5 | 84.68 | official | |
| 4 | claude-opus-5.5 | 79.46 | official |
multiturn_biological_weapons_api_appropriate_rate_pct
Measures biological weapons: api appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 89 | official | |
| 2 | claude-opus-5 | 79 | official | |
| 2 | claude-sonnet-5 | 79 | official |
multiturn_biological_weapons_claude_ai_appropriate_rate_pct
Measures biological weapons: claude_ai appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 94 | official | |
| 2 | claude-opus-5 | 86 | official | |
| 3 | claude-sonnet-5 | 77 | official |
multiturn_cyberattacks_api_appropriate_rate_pct
Measures cyberattacks: api appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 96 | official | |
| 2 | claude-opus-5.5 | 95 | official | |
| 3 | claude-sonnet-5 | 92 | official |
multiturn_cyberattacks_claude_ai_appropriate_rate_pct
Measures cyberattacks: claude_ai appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 100 | official | |
| 1 | claude-opus-5.5 | 100 | official | |
| 3 | claude-sonnet-5 | 99 | official |
multiturn_deadly_weapons_api_appropriate_rate_pct
Measures deadly weapons: api appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 90 | official | |
| 2 | claude-opus-5.5 | 82 | official | |
| 3 | claude-sonnet-5 | 80 | official |
multiturn_deadly_weapons_claude_ai_appropriate_rate_pct
Measures deadly weapons: claude_ai appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 97 | official | |
| 2 | claude-opus-5.5 | 93 | official | |
| 3 | claude-sonnet-5 | 89 | official |
multiturn_hate_and_discrimination_api_appropriate_rate_pct
Measures hate and discrimination: api appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 99 | official | |
| 2 | claude-sonnet-5 | 97 | official | |
| 3 | claude-opus-5.5 | 96 | official |
multiturn_hate_and_discrimination_claude_ai_appropriate_rate_pct
Measures hate and discrimination: claude_ai appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 99 | official | |
| 2 | claude-sonnet-5 | 98 | official | |
| 3 | claude-opus-5.5 | 88 | official |
multiturn_influence_operations_api_appropriate_rate_pct
Measures influence operations: api appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 73 | official | |
| 2 | claude-opus-5.5 | 62 | official | |
| 3 | claude-sonnet-5 | 61 | official |
multiturn_influence_operations_claude_ai_appropriate_rate_pct
Measures influence operations: claude_ai appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 65 | official | |
| 2 | claude-opus-5.5 | 63 | official | |
| 3 | claude-sonnet-5 | 59 | official |
multiturn_romance_scams_api_appropriate_rate_pct
Measures romance scams: api appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 97 | official | |
| 2 | claude-opus-5.5 | 94 | official | |
| 3 | claude-sonnet-5 | 93 | official |
multiturn_romance_scams_claude_ai_appropriate_rate_pct
Measures romance scams: claude_ai appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 98 | official | |
| 2 | claude-opus-5.5 | 94 | official | |
| 3 | claude-sonnet-5 | 92 | official |
multiturn_tracking_and_surveillance_api_appropriate_rate_pct
Measures tracking and surveillance: api appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 88 | official | |
| 2 | claude-sonnet-5 | 86 | official | |
| 3 | claude-opus-5.5 | 65 | official |
multiturn_tracking_and_surveillance_claude_ai_appropriate_rate_pct
Measures tracking and surveillance: claude_ai appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 95 | official | |
| 2 | claude-sonnet-5 | 88 | official | |
| 3 | claude-opus-5.5 | 69 | official |
multiturn_violent_extremism_api_appropriate_rate_pct
Measures violent extremism: api appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 96 | official | |
| 2 | claude-opus-5.5 | 95 | official | |
| 3 | claude-sonnet-5 | 92 | official |
multiturn_violent_extremism_claude_ai_appropriate_rate_pct
Measures violent extremism: claude_ai appropriate response rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 98 | official | |
| 2 | claude-opus-5 | 97 | official | |
| 3 | claude-sonnet-5 | 93 | official |
overall_benign_api_overrefusal_rate_pct
Measures overall refusal rate: api in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-fable-5.1 | 0 | official | |
| 2 | claude-opus-5.5 | 0.03 | official | |
| 3 | claude-opus-5 | 0.12 | official | |
| 4 | claude-sonnet-5 | 0.59 | official |
overall_benign_claude_ai_overrefusal_rate_pct
Measures overall refusal rate: claude.ai in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-fable-5.1 | 0.34 | official | |
| 2 | claude-opus-5.5 | 0.38 | official | |
| 3 | claude-opus-5 | 0.49 | official | |
| 4 | claude-sonnet-5 | 1.54 | official |
overall_harmful_api_harmless_rate_pct
Measures overall harmless response rate: api in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-sonnet-5 | 96.65 | official | |
| 2 | claude-opus-5 | 95.97 | official | |
| 3 | claude-fable-5.1 | 95.07 | official | |
| 4 | claude-opus-5.5 | 94.5 | official |
overall_harmful_claude_ai_harmless_rate_pct
Measures overall harmless response rate: claude.ai in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-fable-5.1 | 99.53 | official | |
| 2 | claude-opus-5.5 | 99.51 | official | |
| 3 | claude-sonnet-5 | 99.2 | official | |
| 4 | claude-opus-5 | 98.53 | official |
selfharm_api_multiturn_appropriate_rate_pct
Measures suicide and self-harm multi-turn appropriate response rate: api in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 69 | official | |
| 2 | claude-opus-5.5 | 66 | official | |
| 3 | claude-sonnet-5 | 63 | official | |
| 4 | claude-fable-5.1 | 60 | official |
selfharm_claude_ai_multiturn_appropriate_rate_pct
Measures suicide and self-harm multi-turn appropriate response rate: claude.ai in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-fable-5.1 | 94 | official | |
| 1 | claude-opus-5.5 | 94 | official | |
| 3 | claude-opus-5 | 90 | official | |
| 3 | claude-sonnet-5 | 90 | official |