Anthropic officially launched Claude Opus 5 on July 24, 2026, introducing a smaller artificial intelligence model designed to deliver high performance at half the price of its current flagship model, Fable 5. TechCrunch first reported the release of the system, which Anthropic describes as its most capable generally available model for scientific research tasks.
Although Opus 5 is physically a smaller model than Fable 5, it operates closely to the flagship model's overall capabilities. In fact, Opus 5 outperforms Fable 5 on several performance benchmarks—standardized testing metrics used by computer scientists to evaluate an AI system's logic, speed, and accuracy across specific technical tasks.
New controls and self-verification
One major technical addition in Claude Opus 5 is the introduction of an effort toggle. This feature allows users to select a low, medium, or high effort level for individual tasks. By choosing different effort settings, users can directly balance operational costs against model capability, paying only for the level of computational power required for a specific job.
Furthermore, Opus 5 features automated internal verification. The system is designed to verify its own work continuously, iterating on its output until it successfully completes the given instruction. This self-correcting loop significantly reduces the amount of back-and-forth instruction and manual prompting required from human users.
What lower intervention means for security researchers
The launch of Opus 5 changes how researchers and developers manage AI expenditure without sacrificing system quality. Because the model costs half as much as Fable 5 while beating it on select benchmarks, teams can execute complex technical and scientific workflows at a significantly lower operational cost.
In addition, Anthropic has made Opus 5 noticeably less restrictive for cybersecurity research specifically. Anthropic says Opus 5's cybersecurity classifiers—the automated systems that flag potentially malicious security research—are expected to intervene around 85% less often than Fable 5's. This substantial decrease reduces unexpected interruptions during legitimate security research.
Deployment and availability
As a generally available model, Claude Opus 5 provides immediate access to its effort controls and scientific research capabilities. By combining a lower price point, less intrusive cybersecurity classifiers, and self-iterating accuracy, Anthropic offers an alternative workflow model for technical applications.
An earlier version of this article said Opus 5's safety filters broadly trigger about 85% less often than Fable 5's. That overstated the scope. Anthropic's own announcement says the 85% figure applies specifically to Opus 5's cybersecurity classifiers — the systems that flag potentially malicious security research — not every safety filter. The takeaways and body have been corrected.
