Anthropic's Opus 5 focuses on token efficiency, not a capability leap
The new Opus 5 model offers performance comparable to Fable at about half the cost, but lags in cybersecurity exploitation due to training decisions.
Anthropic has released Opus 5, the latest update to its popular model series. Unlike previous jumps such as Opus 4.5, this version emphasizes token efficiency over raw capability leaps. Benchmarks provided by Anthropic, including Frontier-Bench and DeepSWE, show Opus 5 performing at a similar level or slightly ahead of the Fable model for coding tasks, but at roughly half the cost. The increments are iterative rather than revolutionary.
However, in cybersecurity, Opus 5 falls behind. Anthropic deliberately omitted cutting-edge training in this area, so the model lags significantly behind Fable and Mythos in exploiting vulnerabilities. It is relatively good at finding them, but “substantially behind Mythos 5 on the exploitation of those vulnerabilities,” according to the company.
Additionally, Opus 5 lacks some of the controversial protections Fable had, such as retaining data for 30 days for review in case of an incident. These choices reflect Anthropic's strategy to provide a cost-effective, specialized tool rather than a one-size-fits-all solution.
