Anthropic’s Opus 5 promises near top model performance for routine work while lowering price and easing guardrails. The surprise tradeoffs could reshape enterprise AI adoption.
In San Francisco on Friday, Anthropic unveiled Opus 5 – a new AI model that, according to the company, comes close in capabilities to the more powerful Fable 5 at a substantially lower price.
The San Francisco-based lab said the new Claude AI model is well suited for everyday office tasks and programming.
In a Reuters interview, Anthropic’s Chief Product Officer Dianne Penn said the release, which proved more efficient than the May Opus 4.8, reflects the rapid pace of development.
We build and continue to steadily provide leading artificial intelligence capabilities and make them as accessible as possible with every generation of models
– Dianne Penn
In trials, Opus 5 demonstrated a reduced ability to exploit cyber vulnerabilities compared with the company’s top model, so its safeguards are less stringent than those of Fable 5. Opus 5 is also less prone to abuse than other current models from the company, the startup reports.
In June, Fable 5 was temporarily unavailable due to concerns in the United States that its capabilities could be used for foreign military intelligence.
According to Penn, users should choose Opus 5 for the value-for-money, and Fable 5 for long-running, highly autonomous projects.
When asked about Moonshot’s open-weight Kimi K3 model from China, which the United States accuses of using others’ work, Penn said that it is not yet clear how open-weight models generally perform in complex real-world tasks that users entrust Claude with.
Open-weight models let users download, run, and customize the AI’s “virtual brain,” unlike private models.
Opus 5 expands Anthropic’s lineup and underscores the company’s focus on the efficiency and accessibility of AI technologies.