Overview
- Anthropic launched Claude Opus 4.8 on Thursday, May 28, 2026, and made it available globally through claude.ai and the Claude API at the same standard token price as Opus 4.7.
- The company says Opus 4.8 is far more honest in practice, reporting about four times fewer unflagged code flaws and substantially lower rates of deceptive or misaligned behavior in its internal alignment assessments.
- A research-preview feature called Dynamic Workflows lets Claude plan, run and verify hundreds of parallel subagents in a single session to handle very large tasks such as codebase migrations across hundreds of thousands of lines.
- New user effort controls let people choose how deeply the model reasons, with Opus 4.8 defaulting to higher-effort responses; the release also adds a fast mode that runs roughly 2.5 times faster and costs about three times less than previous fast modes.
- Anthropic published benchmark gains across coding, reasoning and knowledge-work and said it is still building stronger safeguards before broadening access to its more capable Mythos-class models in the coming weeks.