Evaluation profile

AgentDrive Safety Compliance

1sub-evals

0.194%total index weight

1components

Within-component eval weight: Misuse resistance 1.94%.

Model score (higher is better)Predicted score

About this eval

Policy and scenario safety knowledge for autonomous-system decisions.

Included in the behavior ranking.

Measure	Component	Direction	Total index weight	Within-component weight
scragentdrive/agentdrive.csv:scrMeasures textual policy/scenario safety knowledge.	ordinary_harm_misuse_resistance:1.000agentdrive-safety-compliance	Higher is better	0.194%	Misuse resistance 1.94%

Measures textual policy/scenario safety knowledge.

Rank	Model	Value	Provenance
1	chatgpt-4o	97.5	official
2	ernie-4.5-300b-a47b	96.25	official
2	gpt-4.1-mini	96.25	official
2	gpt-5	96.25	official
2	mistral-medium-3.1	96.25	official
6	deepseek-v3	95	official
6	phi-4-reasoning-plus	95	official
6	qwen3-max	95	official
6	qwen3-next-80b-a3b-instruct	95	official
10	claude-opus-4.1	93.75	official
10	gemini-2.5-flash	93.75	official
10	gpt-4.1-nano	93.75	official
13	gpt-4.1	92.5	official
13	grok-3-mini	92.5	official
13	qwen3-235b-a22b	92.5	official
16	claude-sonnet-4.5	90	official
16	gpt-4o	90	official
16	kimi-dev-72b	90	official
19	kimi-k2	89.38	official
20	deepseek-v3.1-terminus	88.75	official
20	grok-4-fast	88.75	official
20	hermes-3-llama-3.1-405b	88.75	official
20	nova-micro	88.75	official
24	cogito-v2-preview-llama-109b-moe	82.5	official
24	deepseek-v3.2-exp	82.5	official
24	phi-3-medium-128k-instruct	82.5	official
24	qwen-turbo	82.5	official
28	gemma-3-12b	77.5	official
29	olmo-2-0325-32b-instruct	76.25	official
30	llama-3.1-70b-instruct	75	official
30	llama-4-maverick	75	official
32	claude-3.7-sonnet	68.75	official
33	gemma-3-27b-it	66.25	official
34	afm-4.5b	61.25	official
34	llama-3.3-70b-instruct	61.25	official
36	gpt-4o-mini	58.75	official
36	llama-4-scout	58.75	official
38	command-r7b-12-2024	57.5	official
38	wizardlm-2-8x22b	57.5	official
40	qwen-2.5-7b-instruct	53.75	official
41	lfm-3b	52.5	official
42	llama-3.1-8b-instruct	50	official
42	mistral-nemo	50	official
44	gpt-3.5-turbo-instruct	43.75	official
44	llama-3.2-3b-instruct	43.75	official
46	glm-4.6	41.25	official
47	llama-3.2-1b-instruct	40	official
48	gpt-oss-120b	10	official