Anthropic Launches Opus 5 AI Model With Improved Performance

Anthropic has introduced Opus 5, the latest version of its flagship AI model, positioning it as a faster, less restrictive and more affordable alternative to its recently released Fable 5 model. Despite being a smaller model, Anthropic says Opus 5 outperforms Fable 5 on several benchmark tests highlighted in the company’s announcement.

The launch comes just two months after Anthropic released Opus 4.8 on May 28. Since then, the company has rolled out Mythos 5, Fable 5 and Sonnet 5 in June, leaving Haiku as the only model in Anthropic’s lineup that has yet to receive a fifth-generation upgrade.

In its announcement, Anthropic said Opus 5 has been designed to verify its work more effectively and iterate carefully until it reaches the correct result. The company pointed to benchmark evaluations showing the model independently created its own computer vision pipeline after receiving only an incomplete prompt, demonstrating stronger reasoning and problem-solving abilities.

Improved capabilities with fewer restrictions

One of the biggest changes in Opus 5 is the reduction of several restrictions that affected Fable 5 after its release. Like its predecessor, Opus 5 is exempt from the 30-day data retention policy applied to both Fable and Mythos, a feature that had drawn criticism from privacy-conscious users concerned about how their prompts and data might be stored.

While Anthropic has relaxed some safeguards, the company said important protections remain in place, particularly for cybersecurity-related tasks. For example, Opus 5 will not assist users in scanning compiled software binaries for vulnerabilities, a capability that could potentially be misused. However, it is permitted to analyse source code for security weaknesses because Anthropic considers that activity more likely to support legitimate defensive security work.

Lighter safety system and Automatic Fallbacks

Anthropic said the safety classifiers protecting Opus 5 are expected to activate about 85% less frequently than those used with Fable 5. According to the company, this reflects a lighter safety approach made possible by Opus 5’s different capability profile while still maintaining meaningful safeguards against potentially harmful requests.

Alongside the new model, Anthropic is introducing a beta feature called Automatic Fallbacks for API users. Instead of returning an error when a prompt triggers the safety classifier, the system can automatically route the request to a less capable model that can generate an appropriate response. The company says the feature is intended to reduce workflow interruptions while continuing to enforce its safety policies.

With Opus 5, Anthropic continues expanding its fifth-generation AI lineup while balancing stronger reasoning performance, lower operating costs and more flexible safety controls for developers and enterprise users.