Claude Opus 5 succeeds Opus 4.8.

Claude Opus 5 eases cyber security filters. Anthropic now allows code vulnerability research without blocking requests.

The successor to Opus 4.8, Claude Opus 5 is available across all Anthropic platforms and retains the same pricing—five dollars per million input tokens and twenty-five dollars per million output tokens—under the identifier claude-opus-5 on the API side. It becomes the default model for the Max plan. Like its predecessor, it offers an effort setting that balances processing depth against token savings, as well as a fast mode billed at twice the base rate, running at approximately two and a half times the default speed.

The most detailed part of the release focuses on safeguards. Opus 5's cyber classifiers are calibrated less restrictively than those of Fable 5: vulnerability scanning in source code remains permitted, while binary analysis, penetration testing, and exploit generation are blocked. Anthropic estimates that these classifiers will trigger approximately 85% less frequently than on Fable 5. In Claude.ai, Claude Code, and Claude Cowork, a flagged query defaults to falling back to Opus 4.8.

Additionally, a cyber verification program grants pre-approved enterprises and researchers access to a version with relaxed restrictions. On the biology front, queries blocked on Fable 5 are now routed to Opus 5 rather than Opus 4.8. The company states that it deliberately did not train the model on cyber tasks and positions it behind Mythos 5 in both biology research and offensive cybersecurity, considering that it does not push the frontier of dual-use capabilities.

Two beta features accompany the release. Developers can modify the tools accessible to the model mid-conversation without invalidating the prompt cache. An automatic fallback mechanism also allows queries flagged by classifiers—on both Opus 5 and Fable 5—to be routed to another model rather than being blocked.