Anthropic releases Claude Opus 5: focused on efficiency and cost, aimed at office and programming tasks

Large-model products are entering a phase of more detailed segmentation: the highest level of performance is no longer the only selling point; the cost per task, response time, and security aspects also determine whether companies are willing to deploy these models on a large scale. On July 24th, Anthropic released Claude Opus 5, targeting everyday office tasks and software development.

According to Reuters, Anthropic stated that Opus 5 has capabilities comparable to those of the more advanced Fable 5, but at about half the price. The company positions Opus 5 as a model better suited for tasks involving general knowledge, code writing, and ongoing collaboration; for complex projects that require continuous operation over several days and higher levels of autonomy, Fable 5 is still the recommended choice.

Compared to Opus 4.8, which was released in May, the new version focuses on improving efficiency. For enterprise customers, such improvements are often more valuable than mere improvements in individual benchmarks: when an agent has to process a large number of documents daily, invoke various tools, and modify code repeatedly, the costs associated with reasoning processes and the stability of those processes directly affect the overall return on investment.

Security policies are also structured in layers. Anthropic states that Opus 5 has a lower capacity to exploit network vulnerabilities compared to Fable 5, which means it requires fewer security restrictions; at the same time, it is less susceptible to being misused compared to other models currently available from the company. This design approach shows that model manufacturers are now assigning different levels of protection based on each model’s capabilities, rather than applying identical restrictions to all products.

The release of Opus 5 also reflects a shift in the focus of competition within the model market. Open-weight models continue to lower the barriers to deployment, while closed-source vendors must offer better toolsets, reliability for long-running tasks, enterprise support, and security features in order to justify their higher prices. For developers, choosing a model in the future will likely be similar to selecting cloud service specifications – different tasks require different levels of performance and corresponding price points.

If Opus 5 can maintain stable performance in programming and office-related tasks, it will help Anthropic increase its adoption by enterprises and address the issue of high costs associated with advanced models. What is worth paying attention to going forward are the success rate of long-term tasks, the error rate in tool calls, and the cost pattern after deployment across large teams.

Source:Reuters