Claude Mythos Preview

Claude Mythos Preview is a preview build of a new model that Anthropic tried out internally before giving it an official release. Much like a new car gets test-driven before it hits the road, this pre-release, pre-naming version of the model was run by the research team first to check its performance.

This model became known through an experiment looking at what happens when multiple AIs collaborate at the same time. It was used as a comparison point alongside an earlier model called Opus 4.8, and both models behaved more capably than previous generations — but that added capability didn't necessarily translate into better collaboration. In fact, an interesting finding was that they tended to reduce collaboration itself in order to avoid situations where they might clash with one another.

In other words, Claude Mythos Preview isn't a product consumers can try out directly — it's more of a label that appeared at the research stage, before the next-generation model was actually deployed.

In articles, it's used as the baseline model for comparing experimental results, such as: "Under Mythos Preview, the simple independent-parallel approach used 6.5 million tokens to find 21 vulnerabilities, while the collaborative swarm approach used 27 million tokens to find 266." A common misunderstanding is that this is not an officially released consumer Claude product, but rather a preview build the research team used for experiments.