Anthropic launches Opus 5
On Friday, Anthropic announced the release of Opus 5, its latest heavyweight AI model, which is cheaper and less restrictive than Fable 5 while outperforming it on several benchmarks.
Anthropic has released Opus 5, the newest version of its flagship heavyweight AI model, arriving just two months after Opus 4.8 (released May 28). Though smaller than Fable 5, the model is both cheaper and subject to fewer restrictions, making it preferable for most use cases. According to the company, Opus 5 outperforms Fable 5 on a number of benchmarks included in the announcement.
In a blog post, Anthropic emphasized that Opus 5 is "much stronger at verifying its work and iterating carefully until it succeeds," citing an example where the model wrote its own computer vision pipeline in response to an incomplete prompt.
Unlike Fable 5 and Mythos 5, Opus 5 is not covered by the 30-day data retention policy, addressing privacy concerns. However, it still includes safety safeguards, particularly for cybersecurity tasks such as exploit generation and penetration testing. For instance, Opus 5 cannot scan for vulnerabilities in a software binary but may search for them in source code, as the latter is more likely used for defensive purposes.
Anthropic expects these classifiers to engage 85% less often for Opus 5 compared to Fable 5. Additionally, a new beta feature called Automatic Fallbacks allows API users to automatically route requests to a less powerful model when safety classifiers trigger, providing a functional response instead of an error.
