The lab in San Francisco stated that the new Claude AI model is particularly well-suited for everyday office and programming tasks.
In a Reuters interview, Anthropic’s product leader Dianne Penn mentioned that the release, which improves upon May’s Opus 4.8, showcases the rapid pace of innovation.
“We’re constructing and continue to reliably provide frontier intelligence, making it as accessible as possible with each model iteration,” said Penn.
In testing, Opus 5 was found to be less effective at exploiting cyber vulnerabilities compared to Anthropic’s premium AI, which resulted in its safeguards being less stringent than those of Fable 5. The startup also noted that Opus 5 is less prone to misuse than its other models.
After its launch in June, Anthropic’s Fable 5 faced temporary unavailability due to U.S. concerns regarding its potential use in foreign military intelligence.
Penn advised that users should choose Opus 5 for cost-effectiveness and Fable 5 for “days-long, highly autonomous projects.”
When asked about the Kimi K3 “open” model from China’s Moonshot—which the U.S. accused of benefiting from Anthropic’s work—Penn indicated that it “remains to be seen” how these open-weight models perform on complex real-world tasks that users request Claude to handle.
Open-weight models enable users to download, run, and customize the AI’s virtual brains, unlike proprietary models.
