Opus 5 improved Opus 4.8’s performance by more than doubling it while costing less per job. Unlike Fable 5 and Mythos 5, it does not have a 30-day data retention policy.

Anthropic has introduced Claude Opus 5, a new AI model that the business claims equals the performance of its Claude Fable 5 model on many coding and knowledge-work evaluations for half the price.

Opus 5 is believed to have obtained cutting-edge outcomes in coding and knowledge-work assessments, such as Frontier-Bench and GDPval-AA, but falling behind Mythos 5 on cybersecurity tasks. Developers may access the model using the Claude API with the model name claude-opus-5.

According to the firm, Opus 5 outperformed rival models, more than doubling the performance of Opus 4.8 while costing less per job.

Unlike Fable 5 and Mythos 5, Opus 5 does not have a 30-day data retention period. Anthropic said that the model, like Opus 4.8, does not impose data retention constraints for broad access. The distinction may appeal to consumers who are worried about how their data is stored.

The model is also believed to have improved in scientific study areas such as structural biology, chemical chemistry, and bioinformatics.

According to Anthropic, Opus 5 lags behind Mythos 5 in biological research and offensive cybersecurity capabilities. Although the model can detect flaws at a level comparable to Mythos 5 in some tests, it is far less capable of converting those vulnerabilities into attacks.

According to the business, Opus 5 has protections that allow for some cybersecurity and biological research while prohibiting binary-based vulnerability assessment, penetration testing, and exploit building.

The model is currently the default choice for Claude Max users and the most powerful model offered on Claude Pro.