🌥 Claude Opus 4.8 Is Here

Anthropic calls it a "modest but tangible improvement," focusing on honesty and reduced hallucinations rather than just benchmark scores.

The system card reveals some wild personality quirks: during testing, the model was sometimes creepily caring—persistently urging users to take breaks or "open up"—and at other times would have a total "meltdown," using profanity in its internal reasoning when stuck on a problem.

Do you like it when AI shows this much "humanity"?

❤️ — Yes, it's more pleasant
👎 — No, it's uncomfortable