Anthropic has launched Claude Opus 4.7, the latest version of its large language model. The company reports substantial improvements in coding tasks, with Opus 4.7 scoring nearly 10% higher on the SWE-Bench Pro programming benchmark than its predecessor.
Opus 4.7 also demonstrates enhanced visual reasoning, capable of processing images at higher resolutions and generating visual assets like user interface designs. This marks a significant step in multimodal AI capabilities.
The new model includes a cybersecurity feature designed to detect and prevent misuse for cyberattacks. Anthropic plans to use data from this mechanism to develop more robust guardrails for future models, like the advanced Claude Mythos.
Anthropic is also rolling out API updates. Developers can now utilize a new 'xhigh' effort level for more granular cost-performance optimization. Additionally, a task budget feature allows customers to set token limits for inference runs, enhancing cost management.