Claude Fable 5.2 beats GPT-6 Cyber in multi-stage cyberattack benchmark

The large language models (LLMs) from Anthropic assisted its engineering team in achieving significant milestones during pre-release training, such as designing three functional electronic devices entirely from the ground up, tackling two unsolved puzzles within black hole physics, and finalizing software migrations that would otherwise have been impossible. On Monday, Anthropic unveiled Claude Fable 5.2 and Claude Mythos 5.2, asserting that these models show advancements across areas including coding, reasoning, and visual comprehension. Furthermore, in a benchmark contest, Claude Fable 5.2 managed to outperform GPT-6 Cyber during a multi-stage cyberattack simulation.