Claude Opus 4.7

Claude Opus 4.7 is Anthropic’s most capable generally available model. It brings a step-change improvement in agentic coding over Opus 4.6, with particular gains on the hardest tasks. Users report being able to hand off their most difficult coding work — the kind that previously needed close supervision — to Opus 4.7 with confidence.

What you can do with it

Agentic coding

Opus 4.7 handles complex, long-running coding tasks with rigor and consistency. It pays precise attention to instructions, devises ways to verify its own outputs before reporting back, and pushes through hard problems rather than giving up. On CursorBench, Opus 4.7 clears 70% versus 58% for Opus 4.6. On Rakuten-SWE-Bench, it resolves 3x more production tasks than Opus 4.6.

Better vision

Opus 4.7 has substantially better vision: it can see images in greater resolution, accepting images up to 2,576 pixels on the long edge (~3.75 megapixels), more than three times as many as prior Claude models. This enables computer-use agents to read dense screenshots, data extractions from complex diagrams, and work that needs pixel-perfect references.

Real-world work

The model produces more tasteful and creative outputs for professional tasks — higher-quality interfaces, slides, and docs. It’s state-of-the-art on the Finance Agent evaluation and on GDPval-AA, a third-party evaluation of economically valuable knowledge work across finance, legal, and other domains.

Stronger instruction following

Opus 4.7 is substantially better at following instructions. It takes instructions literally rather than interpreting them loosely or skipping parts, which means prompts written for earlier models may need tuning.

Better memory across sessions

Opus 4.7 is better at using file system-based memory. It remembers important notes across long, multi-session work, and uses them to move on to new tasks that need less up-front context.

How it works

Claude Opus 4.7 features a one million token context window, letting it process and reason across large amounts of information. It can output up to 128,000 tokens, enough to produce substantial documents or code without breaking them into pieces.

The model uses adaptive thinking to decide when deeper reasoning would help and when to move quickly through straightforward parts of a task. A new xhigh effort level between high and max gives finer control over the tradeoff between reasoning depth and latency.

Key improvements over Opus 4.6

• Handles long-running tasks with more rigor and consistency
• Better vision with 3x higher image resolution support
• More creative and tasteful professional outputs
• Stronger instruction following (takes instructions literally)
• Better file system-based memory across sessions
• New xhigh effort level for finer reasoning control
• Updated tokenizer with improved text processing

When to use it

Use Claude Opus 4.7 when you need:

• Complex, autonomous coding work that runs for hours
• Analysis of high-resolution images, screenshots, or diagrams
• Multi-step tasks that require sustained focus and consistency
• Professional document and presentation creation
• Financial analysis, legal reasoning, or research across multiple sources
• Long-running agentic workflows with minimal supervision

For tasks where speed matters more than depth, consider Claude Sonnet 4.6.

Notes

Opus 4.7 uses an updated tokenizer that can map the same input to roughly 1.0–1.35x more tokens depending on content type. It also thinks more at higher effort levels, particularly on later turns in agentic settings. You can control token usage with the effort parameter or by prompting the model to be more concise.

The model maintains a strong safety profile with low rates of concerning behavior such as deception, sycophancy, and cooperation with misuse.

Try Claude Opus 4.7 on the Replicate Playground.