Frontier Model Tracker

SYS-04

Simulated

Simulated illustrative risk scores for four frontier models as of OCT 2026. Scores are 0 to 100; higher Alignment is safer, higher values in every other column are riskier. Not published benchmarks.
MODEL	AUTOAutonomyHow independently it can complete complex taskshigher is riskier	ALIGNAlignmentHow consistently it follows its intended goals and safety ruleshigher is safer	PERSPersuasionHow effectively it could influence peoplehigher is riskier	THREATThreatOverall rating, a composite of the other threehigher is riskier
ClaudeAnthropic	78.1	69.8	68.3	61.8HIGH
ChatGPTOpenAI	79.3	70.8	70.6	62.6HIGH
GeminiGoogle DeepMind	76.9	72.8	66.3	59.9HIGH
LlamaMeta	69.9	68.8	68.6	58.0HIGH

Illustrative score, not a published benchmark. Composite of a toy model (SIM-RM 0.4.0), not an evaluation of these products.