SAN FRANCISCO, July 24 - Anthropic on Friday introduced Opus 5, the latest addition to its Claude lineup. The company said the model approaches the performance of its more capable sibling, Fable 5, while being offered at about half the price.
Anthropic described Opus 5 as particularly well suited to everyday office tasks and computer programming assignments. In an interview, product leader Dianne Penn said the model’s launch represented another step in rapid development, noting that Opus 5 is more efficient than the May release, Opus 4.8.
"We’re building and continue to consistently deliver frontier intelligence and bring that as accessibly as possible with every model generation," Penn said.
According to Anthropic’s internal testing, Opus 5 is less capable than the company’s top-tier AI at exploiting cyber vulnerabilities. That relative limitation has informed the company’s approach to safety restrictions: Opus 5 carries less restrictive safeguards than Fable 5 because it showed a lower ability to perform such harmful tasks in tests.
Anthropic also reported that Opus 5 proved harder to trick into misuse compared with its other currently deployed models.
Fable 5, released in June, had a temporary interruption in availability after U.S. authorities raised concerns that its capabilities might be repurposed for foreign military intelligence. Penn advised users to select models based on task scope: Opus 5 is the value choice for routine work, whereas Fable 5 is more appropriate for "days-long, very autonomous projects," she said.
When asked about the Kimi K3 "open" model produced by China-based Moonshot - which U.S. officials have accused of drawing on Anthropic’s work without permission - Penn said it "remains to be seen" how open-weight models broadly will fare when deployed on the complex, real-world projects that customers currently ask Claude to handle.
Anthropic explained that open-weight models differ from proprietary models because they allow users to download, run and customise the underlying model weights - effectively enabling external parties to run and modify the model’s virtual 'brain' locally.
Implications
The company frames Opus 5 as an efficiency-focused alternative intended for frequent, everyday tasks and for users seeking lower-cost compute. At the same time, Anthropic draws a distinction between that value orientation and the higher-capability, more restricted Fable 5 for extended autonomous workloads.