Anthropic today Introduced Opus 5the latest update to a model that has recently become a popular choice for coding and other software development tasks, among other things.
While this is a notable bump for Opus, the agent doesn’t seem like an Opus 4.5-level improvement in encoding performance.
A graph with various benchmarks such as Frontier-Bench and DeepSWE (produced by Anthropic) shows that Opus 5 is about on par with or slightly ahead of Anthropic’s very powerful Fable model for coding tasks. It outperforms Opus 4.8 and OpenAI’s competing GPT-5.6-Sol in almost every task.
Various benchmarks show an iterative increase in performance, but not a radical leap, the main one being that this is a model that offers just shy of the Fable for about half the price.
It’s also worth noting that Anthropic specifically avoids giving Opus 5 advanced training in cybersecurity tasks, so it’s far behind Fable and Mythos in that regard. Anthropic claims that it is relatively good at finding cybersecurity vulnerabilities, but that it is “significantly behind Mythos 5” in terms of decisions made in training the model. exploitation of these weaknesses.”
As such, Opus 5 lacks all of the same controversial safeguards as Fable, such as its policy of retaining data for review for 30 days in the event of an incident.
