Anthropic has launched Claude Opus 5, presenting it as a more capable and cost-efficient successor to Opus 4.8 for coding, research, and complex professional work. Released on July 24, 2026, the model is available across Anthropic’s platforms, serving as the default model for Claude Max and the strongest option on Claude Pro. The launch broadens Anthropic’s fifth-generation lineup with a model intended for advanced everyday use.
Performance and Efficiency
Anthropic said Opus 5 delivers substantially better performance at the same base price as Opus 4.8, with adjustable effort settings that balance intelligence, speed, and token use. On Frontier-Bench v0.1, the company reported that it surpassed competing models and more than doubled its predecessor’s result at a lower cost per task. On CursorBench 3.2, it reportedly approached Fable 5’s peak performance while costing roughly half as much per task.
Coding and Agentic Work
The model is designed to verify its output, revise its approach, and persist through difficult assignments with less human intervention. In one evaluation, Opus 5 built a computer vision pipeline to extract geometry from a drawing it could not directly view, then reconstructed a three-dimensional machine part. Anthropic also said it found the root cause of a software bug, fixed a missed edge case, and created validation tools when external testing resources were unavailable.
Knowledge Work and Research
Beyond coding, Anthropic is promoting Opus 5 for business automation, computer use, financial analysis, and scientific research. The company reported leading cost-adjusted results on evaluations including ARC-AGI 3, AutomationBench, OSWorld 2.0, GDPval-AA, and DeepSearchQA, although several claims rely on Anthropic’s own testing. It also recorded gains across structural biology, organic chemistry, and bioinformatics, particularly in molecular structure inference and protein-related tasks.
Enterprise Applications
Early-access customers reported improvements in debugging, numerical reasoning, legal analysis, financial modeling, genomics, code review, and visual content creation. Companies including Cognition, Cursor, Zapier, Box, JetBrains, and Lovable said the model handled longer, less clearly defined workflows more consistently than earlier Opus releases. Their feedback supports Anthropic’s effort to position Opus 5 as an operational engine for agents completing multi-step development and enterprise tasks.
Safety and Safeguards
Anthropic described Opus 5 as its most aligned model to date, citing lower rates of deceptive and reckless behavior in pre-deployment audits. The model remains behind Mythos 5 in advanced biological research and offensive cybersecurity, and Anthropic said it did not train Opus 5 specifically for high-risk cyber capabilities. Its safeguards allow defensive source-code analysis but restrict binary scanning, penetration testing, and exploit generation, with certain blocked requests routed to Opus 4.8.
Pricing and Availability
Claude Opus 5 is available through Claude products and the API as claude-opus-5, priced at $5 per million input tokens and $25 per million output tokens. A Fast mode runs about 2.5 times faster at twice the base price, while new beta features support mid-conversation tool changes and automatic model fallbacks. Unlike Fable 5 and Mythos 5, Opus 5 carries no general-access data-retention requirement, which may appeal to privacy-sensitive organizations.
Claude Opus 5 strengthens Anthropic’s competitive position by emphasizing practical performance, predictable pricing, and reliability rather than benchmark leadership alone. Its stronger coding capabilities, broader knowledge-work performance, adjustable effort levels, and comparatively flexible access could make it Anthropic’s most broadly useful premium model for developers and enterprises. Its lasting impact will depend on whether the reported gains hold across production workloads while safeguards remain effective as organizations give the model greater autonomy.