# AI safety evaluation

OpenAI is rolling out GPT-6 Sol and GPT-6 Luna to ChatGPT users on free and paid plans worldwide, replacing GPT-5.6 Sol and GPT-5.6 Luna there, while the same-named models in Codex and ChatGPT Work remain the September versions. Against the GPT-5.6 pair in ChatGPT, GPT-6 showed stronger resistance to jailbreaks, including attacks that adapt across multiple turns, and reductions in dishonesty, deception and circumvention of guardrails. Under the Preparedness Framework, both October models are treated as High capability in Cybersecurity and in Biological and Chemical domains, and neither reaches the High threshold in AI Self-Improvement.

OpenAI is rolling out GPT-6 Sol and GPT-6 Luna to ChatGPT users on free and paid plans worldwide, replacing GPT-5.6 Sol and GPT-5.6 Luna there, while the same-named models in Codex and ChatGPT Work remain the September versions. Against the GPT-5.6 pair in ChatGPT, GPT-6 showed stronger resistance to jailbreaks, including attacks that adapt across multiple turns, and reductions in dishonesty, deception and circumvention of guardrails. Under the Preparedness Framework, both October models are treated as High capability in Cybersecurity and in Biological and Chemical domains, and neither reaches the High threshold in AI Self-Improvement.