Claude Opus 4.7 Arrives With Sharper Vision, Self-Checking Code, and Real-Time Cyber Guardrails
News Summary
Anthropic officially released Claude Opus 4.7 on April 16, 2026 (PDT), marking the latest upgrade to its flagship AI model lineup. The new model delivers significant improvements in advanced software engineering, high-resolution vision, and autonomous task execution, while maintaining the same pricing as its predecessor. Claude Opus 4.7 is now broadly available through Anthropic's own products, its API, and major cloud platforms including Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
Core Capabilities and Performance Improvements
Claude Opus 4.7 represents a notable improvement over Opus 4.6 in advanced software engineering, with particular gains on the most difficult tasks. The update brings a 13% lift on coding benchmarks and three times more production tasks resolved, along with high-resolution vision support up to 3.75 megapixels and a new tokenizer. In terms of self-verification, the model devises ways to verify its own outputs before reporting back, and handles complex, long-running tasks with rigor and consistency.
Anthropic introduced a new "xhigh" effort level — positioned between the existing "high" and "max" settings — giving users finer control over the tradeoff between reasoning depth and response latency on demanding problems. The company is also piloting a new "task budgets" system that gives developers greater control over how Claude allocates reasoning across longer workflows.
Vision and Multimodal Upgrades
Maximum image resolution jumped from 1,568 pixels on the long edge to 2,576 pixels, roughly tripling the visual capacity of previous Claude models. This improvement brings practical benefits across several domains: the model's coordinates now map 1:1 with actual pixels for computer use and screenshot understanding, eliminating the scale-factor calculations that were previously required. Higher resolution also allows the model to read smaller text and finer details in scanned documents, slides, and diagrams.
Memory and Long-Horizon Task Handling
Opus 4.7 is better at using file system-based memory, remembering important notes across long, multi-session work and using them to move forward on new tasks that, as a result, need less up-front context. The model delivers stronger multi-step task performance and more reliable agentic execution, with meaningful improvement in long-horizon reasoning and complex, tool-dependent workflows.
Cybersecurity Safeguards and Responsible Deployment
Opus 4.7 is the first Claude model to ship with real-time cybersecurity safeguards that automatically detect and block requests indicating prohibited or high-risk cybersecurity uses. During training, Anthropic experimented with efforts to "differentially reduce" Opus 4.7's cyber capabilities. Security professionals wishing to use the model for legitimate purposes such as vulnerability research and penetration testing are invited to apply through a new Cyber Verification Program. Anthropic has stated that insights gained from deploying these safeguards in the real world will inform the path toward eventually making its more advanced Mythos-class models more broadly accessible.
Benchmark Position and Model Landscape
In benchmark comparisons, Opus 4.7 outperforms Opus 4.6, as well as competing models from other major AI providers, across a number of key metrics. However, it still falls short of Anthropic's own Claude Mythos Preview model, which has only been released to a select group of technology and cybersecurity companies. Opus 4.7 also achieved state-of-the-art results on GDPval-AA, a benchmark that measures economically valuable knowledge work in finance and legal domains.
Pricing, Availability, and Migration Notes
Pricing remains unchanged from Opus 4.6, at $5 per million input tokens and $25 per million output tokens. The model is available across all Claude products and the API, as well as Amazon Bedrock, Google Cloud's Vertex AI, and Microsoft Foundry. Anthropic notes that Opus 4.7 uses an updated tokenizer that can map the same input to roughly 1.0–1.35 times more tokens depending on content type, a tradeoff developers should account for when migrating from Opus 4.6. Anthropic has established a more predictable two-month cadence for Opus model upgrades, with Opus 4.7 arriving two months after Opus 4.6, which itself followed two months after Opus 4.5.