GPT-5.5 matches heavily hyped Mythos Preview in new cybersecurity tests

OpenAI’s GPT-5.5 matched the heavily hyped Mythos Preview in independent cybersecurity benchmarks, proving that iteration can outperform flashy launches. In tests against 1,200 real-world exploit scenarios, GPT-5.5 achieved a 94.3% detection rate, identical to Mythos Preview's score but at half the computational cost. This outperforms #2's previous model by 12% in false positive reduction. The benchmark also showed GPT-5.5 is 30% faster than the average large language model at analyzing zero-day threats. Rather than chasing publicity, OpenAI focused on refinement, which consistently delivers results that outclass more heavily marketed competitors.