The surprise of the new Claude Opus 4.8 isn’t that it’s (a little) better. The surprise is its “I only know that I know nothing”

Anthropic released Claude Opus 4.8 just 41 days after Opus 4.7, with modest benchmark gains over prior Claude, GPT-5.5, and Gemini 3.1 Pro results.

The bigger focus is reliability: Anthropic says the model is more honest about uncertainty, catches its own mistakes more often, and is better aligned.

Claude Code also gets preview Dynamic Workflows, letting users run large parallel agent sessions for complex codebase analysis and migrations.

Anthropic says even stronger public models are coming later, but they will wait on additional cybersecurity safeguards first.