Released on April 16, 2026, Claude Opus 4.7 is Anthropic's flagship large language model optimized for high-stakes production environments, advanced software engineering, and long-horizon autonomous agents. It serves as a direct upgrade to the Opus 4.6, offering enhanced reliability in multi-step reasoning and a significant leap in visual processing capabilities. The model is designed to handle ambiguity by making sensible assumptions and self-verifying its outputs before final delivery.
One of the defining features of this version is the introduction of a new xhigh (extra high) effort level. This setting allows users to control the trade-off between reasoning depth and latency, particularly in its "non-reasoning" mode where internal thinking traces are suppressed but the model's internal compute allocation is maximized for precision. This high-effort configuration is recommended for complex coding tasks and agentic workflows where literal instruction following and consistency are prioritized over response speed.
Opus 4.7 features a substantial vision upgrade, supporting resolutions up to 3.75 megapixels (2,576 pixels on the long edge). This 3x increase in detail compared to its predecessors enables more effective data extraction from dense screenshots, architectural diagrams, and intricate financial documents. Additionally, the model utilizes a new tokenizer that improves performance across various languages and codebases, though it may result in higher token counts for certain heavy-text workloads.
Performance-wise, the model records a score of 87.6% on SWE-bench Verified, demonstrating a strong capability for autonomous systems engineering. It maintains a 1-million token context window and a 128,000-token maximum output limit. To help manage costs associated with high-effort processing and long-running loops, Anthropic introduced task budgets, allowing developers to set caps on token spend for individual agentic sessions.