OpenAI’s GPT-6 and DeepSeek V4.1 Show Strong Cyberattack Performance in AI Benchmark

OpenAI’s latest GPT-6 Sol Daybreak model achieved an 80.8 percent success rate in hitting attack objectives within the Vals AI benchmark, which simulates corporate network cyber capabilities, according to The Decoder. This marks a significant milestone in AI-driven cybersecurity simulations.

Meanwhile, the open-weight model DeepSeek V4.1 Flash also demonstrated strong performance, reaching 72.8 percent of attack objectives at just a tenth of the cost compared to GPT-6. The cost efficiency of DeepSeek’s model highlights the growing accessibility of advanced AI tools in cybersecurity testing.

For Japanese markets, these advancements underline the importance of integrating cutting-edge AI into cybersecurity strategies, especially as digital infrastructure expands and cyber threats become more sophisticated.