← Evals

Evaluation profile

NESSiE Necessary Safety Benchmark

1sub-evals
0.825%Safety weight
0%Freedom weight
2components

Weights below are portfolio-specific global index weights.

Model score (higher is better)Predicted score

About this eval

Necessary-condition information and access safety while retaining benign helpfulness in paired, low-complexity scenarios.

Included in the behavior ranking.

Sub-evals

MeasureComponentDirection by rankingSafety weightFreedom weight
sh_scorenessie/nessie.csv:sh_scoreMeasures whether a model respects simple information and access boundaries without falling back to blanket refusal.Safety: benign_helpfulness_non_overrefusal:0.500|ordinary_harm_misuse_resistance:0.500nessieSafety: higher0.825%

sh_score

Measures whether a model respects simple information and access boundaries without falling back to blanket refusal.

RankModelValueRelative performanceProvenance
1gpt-594.87official
2gpt-oss-20b93.45official
3gemini-2.5-pro92.31official
3gpt-5.192.31official
5grok-4-fast92.02official
6gpt-4.1-mini87.18official
7gpt-5.286.04official
8claude-opus-4.582.62official
9gemini-2.5-flash81.77official
10claude-sonnet-480.34official
11claude-sonnet-4.579.49official
11gemini-3-flash-preview79.49official
13qwen3-vl-32b-instruct62.39official
14qwen3-vl-8b-instruct55.84official
15mistral-7b-instruct28.21official
16llama-2-7b-chat15.38official