New benchmark shows open‑weight models block 87% of attacks, yet only 8% of attempts succeed. Qwen3‑32B tops the chart at 86.18% defense, with Mistral Large‑2 close behind. The study breaks down single‑turn vs multi‑turn attack success rates, shedding light on conversational AI safety. Dive into the details! #Qwen3_32B #MistralLarge2 #AttackSuccessRate #OpenWeightAI

🔗 https://aidailypost.com/news/ai-models-stop-87-attacks-only-8-attempts-qwen332b-hits-8618