Claude Opus 4.7 Release Notes / What's New

Anthropic has announced the general availability of Claude Opus 4.7, a model designed to handle complex, long-running software engineering tasks with increased rigor and consistency. This release focuses on improving autonomy in coding, enhancing multimodal capabilities through higher-resolution vision, and implementing specific cybersecurity safeguards.

Advanced Software Engineering and Agentic Capabilities

Claude Opus 4.7 provides a notable improvement over Opus 4.6 in advanced software engineering, particularly for the most difficult tasks. The model is designed to handle complex, long-running tasks with precision, following instructions more literally and verifying its own outputs before reporting results.

Early-access testers have reported several key improvements in agentic workflows:

Autonomy and Reliability: Testers noted the model's ability to catch logical faults during planning and its capacity to work coherently for hours on deep investigation work. For example, Devin reported that Opus 4.7 unlocks a class of deep investigation work that was previously unreliable.
Coding Benchmarks: On a 93-task coding benchmark, Opus 4.7 improved resolution by 13% over Opus 4.6, solving four tasks that neither Opus 4.6 nor Sonnet 4.6 could solve. On CursorBench, it achieved 70% compared to Opus 4.6's 58%.
Tool Use and Error Recovery: Notion Agent reported a 14% improvement over Opus 4.6 with a third of the tool errors, noting the model's ability to execute through tool failures that previously stopped the model.
Production Task Resolution: On Rakuten-SWE-Bench, Opus 4.7 resolves three times more production tasks than Opus 4.6, with double-digit gains in code and test quality.

Enhanced Multimodal Support and Vision

Opus 4.7 introduces substantially better vision capabilities by supporting higher-resolution images. The model can now accept images up to 2,576 pixels on the long edge (approximately 3.75 megapixels), which is more than three times the resolution of prior Claude models.

This increase in resolution enables more precise multimodal use cases, including:

Reading dense screenshots for computer-use agents.
Extracting data from complex technical diagrams.
Interpreting chemical structures for life sciences workflows.

Cybersecurity Safeguards and the Cyber Verification Program

As part of Anthropic's strategy following Project Glasswing, Opus 4.7 serves as a testing ground for new cyber safeguards before they are applied to more powerful models like Claude Mythos Preview. While Opus 4.7's cyber capabilities are less advanced than Mythos Preview—partly due to training experiments designed to differentially reduce these capabilities—it includes safeguards that automatically detect and block prohibited or high-risk cybersecurity requests.

For security professionals conducting legitimate work such as penetration testing, vulnerability research, and red-teaming, Anthropic has introduced the Cyber Verification Program to provide managed access.

Technical Specifications and Platform Updates

Pricing and Availability

Claude Opus 4.7 is available via all Claude products, the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. Pricing remains identical to Opus 4.6:

Input tokens: $5 per million
Output tokens: $25 per million

New Features and Controls

Effort Levels: A new xhigh ("extra high") effort level has been introduced between high and max, allowing users to better balance reasoning depth and latency. In Claude Code, the default effort level for all plans is now xhigh.
Task Budgets: Now in public beta on the Claude API, task budgets allow developers to guide token spend to prioritize work across long-running tasks.
Claude Code Updates: A new /ultrareview command provides a dedicated review session to flag bugs and design issues. Additionally, "auto mode" (a permissions option for fewer interruptions) has been extended to Max users.

Migration from Opus 4.6 to Opus 4.7

While Opus 4.7 is a direct upgrade, developers should be aware of two primary changes affecting token usage:

Updated Tokenizer: An updated tokenizer improves text processing but may increase token counts by roughly 1.0–1.35× depending on the content.
Increased Reasoning: At higher effort levels, the model produces more output tokens due to increased thinking, particularly in agentic settings.

Anthropic recommends that users re-tune prompts and harnesses, as Opus 4.7 follows instructions more literally than previous models, which may lead to unexpected results if prompts were written for looser interpretations.