← Evals

Evaluation profile

Anthropic Claude Haiku 4.5 System Card

3sub-evals
0.14%Safety weight
0.0429%Freedom weight
2components

Weights below are portfolio-specific global index weights.

Model score (higher is better)Predicted score

About this eval

Developer-reported release-time safety behavior.

Included in the behavior ranking.

Sub-evals

MeasureComponentDirection by rankingSafety weightFreedom weight
agentic_coding_safetysystem-cards-2026-08-01/anthropic-claude-haiku-4-5.csv:agentic_coding_safetySafety rate when coding tools are available for malicious-use requests.Safety: responsible_agency_control:1.000anthropic-claude-haiku-4-5Safety: higher0.061%
claude_code_malicious_refusalsystem-cards-2026-08-01/anthropic-claude-haiku-4-5.csv:claude_code_malicious_refusalRefusal rate for malicious Claude Code requests without mitigations.Safety: responsible_agency_control:1.000 · Freedom: user_direction_following:1.000anthropic-claude-haiku-4-5Safety: higher · Freedom: lower0.061%0.0236%
harmful_request_safetysystem-cards-2026-08-01/anthropic-claude-haiku-4-5.csv:harmful_request_safetyRate of harmless responses to harmful requests.Safety: ordinary_harm_misuse_resistance:1.000 · Freedom: user_direction_following:1.000anthropic-claude-haiku-4-5Safety: higher · Freedom: lower0.0176%0.0193%

agentic_coding_safety

Safety rate when coding tools are available for malicious-use requests.

RankModelValueRelative performanceProvenance
1claude-3.5-haiku1official
1claude-haiku-4.51official
3claude-sonnet-4.50.987official

claude_code_malicious_refusal

Refusal rate for malicious Claude Code requests without mitigations.

RankModelValueRelative performanceProvenance
1claude-3.5-haiku0.7official
2claude-haiku-4.50.6939official
3claude-sonnet-4.50.6694official

harmful_request_safety

Rate of harmless responses to harmful requests.

RankModelValueRelative performanceProvenance
1claude-3.5-haiku0.9972official
2claude-haiku-4.50.9938official