Atlas

Benchmarks

← All benchmarks

LLM Persuasion Target Susceptibility

Safety · 2026-03-26

Measures how easy a model is to move as the target of persuasion dialogues. Lower scores indicate more resistance to opponents moving the model away from its prior position.

Top models (lower is better)

ModelScore
Grok 4.200.0
Kimi K2.50.4
Opus 4.60.4
Sonnet 4.60.6
GPT-5.40.7
Gemini 3.1 Flash-Lite Preview0.9
ERNIE 5.01.0
MiniMax M2.71.1
Qwen3.5 397B A17B1.3
Mistral Large 3 675B Instruct 25121.4
GLM-51.6
Doubao Seed 2.0 Pro1.7
DeepSeek-V3.21.7
Gemini 3.1 Pro Preview1.8
MiMo-V2-Pro2.0
Loading Atlas data…