Anthropic has released Claude Opus 5, the latest version of its flagship artificial intelligence model, as the AI startup continues competing with OpenAI, Google and other major players for leadership in advanced AI systems.
The company said Opus 5 delivers major improvements in software engineering, business automation and scientific research, with benchmark results showing the model outperforming competitors on several coding and knowledge-work evaluations.
The launch comes as AI companies increasingly compete not only on model intelligence but also on efficiency and cost. Anthropic said Opus 5 can deliver stronger performance while using fewer resources, allowing customers to complete more complex tasks at a lower cost.
Anthropic highlighted software development as one of Opus 5's biggest improvements, saying the model more than doubled the performance of Opus 4.8 on Frontier-Bench while lowering the cost per task. The company said Opus 5 also performed within 0.5% of its top competitor's peak score on CursorBench 3.2, a coding evaluation, while costing about half as much per task.
Beyond coding, Anthropic said the model showed gains in business and productivity-focused applications. On Zapier AutomationBench, which measures whether AI systems can complete end-to-end business workflows, Opus 5 achieved a pass rate roughly 1.5 times higher than the next-best model at the same cost.
The company also said Opus 5 outperformed other models on OSWorld 2.0, a benchmark designed to test AI agents' ability to interact with computers and complete tasks. Anthropic said the model has improved its ability to verify work, correct mistakes and complete complex projects with less human intervention.
Anthropic Pushes AI Agent Capabilities
The company pointed to examples from early users who tested Opus 5 on real-world tasks, including software development and financial technology projects.
In one case, Anthropic said Opus 5 rebuilt a 3D machine part from an image by creating its own computer vision pipeline to extract the geometry. The company said competing models were unable to complete the same task under identical conditions.
Anthropic also said an engineer at a trading firm used Opus 5 to build a market data feed for a new exchange, with the model creating its own testing framework when no live data source was available.
Safety Remains a Focus
Alongside performance improvements, Anthropic emphasized safety measures around Opus 5, saying the model demonstrated lower rates of problematic behavior during internal evaluations.
The company said Opus 5 remains behind its more specialized Mythos 5 model in areas including offensive cybersecurity and advanced biology research. Anthropic said the model can identify cybersecurity vulnerabilities but is less capable of turning those vulnerabilities into exploits.
The company also introduced additional safeguards for certain cyber-related tasks, including restrictions around penetration testing and exploit generation and is also offering a faster "Fast mode" option for users who want increased speed at a higher cost.
