Anthropic Releases New Claude Opus 4.7

Ronni Holmvig Strøm · 2026-04-17

Anthropic has officially launched Claude Opus 4.7, its most capable generally available model to date.

Anthropic has officially launched Claude Opus 4.7, its most capable generally available model to date.

Released on April 16, the new flagship pushes performance boundaries in advanced software engineering, long-horizon agentic workflows, and high-resolution vision, while keeping pricing unchanged from Opus 4.6.

Positioned as a practical upgrade rather than a full frontier reset, Opus 4.7 arrives amid growing excitement (and some restraint) around Anthropic’s even more powerful but restricted Claude Mythos Preview. For developers, enterprises, and research teams, this release signals faster, more autonomous AI assistance on the hardest real-world tasks.

Key Improvements in Claude Opus 4.7

Anthropic highlights several meaningful gains that make Opus 4.7 especially strong for production environments:

Advanced Software Engineering & Agentic Work. Users can now hand off complex, long-running coding tasks that previously required close supervision. The model shows greater rigor, consistency, and self-verification, it plans more carefully, follows instructions precisely, and checks its own outputs before reporting back.

Vision Capabilities Jump. Opus 4.7 supports significantly higher image resolution (up to 2,576 pixels on the long edge, roughly 3.75 megapixels, more than 3× previous Claude models). This enables better analysis of dense screenshots, technical diagrams, UI mockups, and documents, while also improving the quality of generated slides, interfaces, and visual content.

New Reasoning Controls. A new “xhigh” effort level sits between “high” and “max,” giving finer control over the reasoning depth vs. speed tradeoff. Task budgets and an “ultrareview” feature in Claude Code further support reliable long-running agent work.

1M Context Window. Building on the beta introduced in Opus 4.6, the large context remains available for complex, multi-file projects and extended workflows.

Benchmark Highlights

Opus 4.7 delivers solid gains on key evaluations, particularly those measuring real-world coding and agent performance:

SWE-bench Verified: 87.6% (improved from Opus 4.6) SWE-bench Pro: 64.3% Internal 93-task coding benchmark: +13% resolution rate over Opus 4.6, including four tasks that neither 4.6 nor Sonnet 4.6 could solve Vision & reasoning tasks: Strong improvements in CharXiv, GPQA Diamond, Finance Agent, and Terminal-Bench CursorBench (real-world IDE coding): Reported jumps to ~70% in early tests