GPT-5.6 Sol outperforms Mythos 5 on AISI’s cyber challenge

Source: AISI

They’ve also added GLM 5.2 and DeepSeek V4 Pro. AISI says leading open-weight models are now roughly 4–7 months behind the closed-model frontier, narrowing from 6–10 months through most of 2025. They haven’t benchmarked K3 yet, so it’ll be interesting to see how it changes the gap once AISI tests it.