Claude Opus 4.7: Anthropic’s Sharper, More Reliable Coding and Multimodal Model

AI assistant collaborating with developer at a laptop, surrounded by code snippets and diagrams

Anthropic’s newest release, Claude Opus 4.7, is now generally available. Built as an incremental but meaningful upgrade over Opus 4.6, Opus 4.7 is positioned as a model that improves sustained reasoning, long-running workflows, and high-resolution vision. Anthropic emphasizes that while Opus 4.7 is not as broadly capable as their most advanced Mythos Preview model, it brings tangible gains for software engineering, multimodal tasks, and enterprise agent orchestration, and it will be deployed first as part of their careful rollout strategy.

What Opus 4.7 improves for developers

Anthropic highlights Opus 4.7 primarily as a leap forward for coding and multi-step engineering work. The model demonstrates stronger autonomy on complex, long-running tasks: it plans more robustly, detects logical faults earlier in the planning phase, and follows instructions more strictly. In internal benchmarks and partner evaluations, Opus 4.7 shows higher task-resolution rates and fewer tool errors than Opus 4.6, often completing workflows that previously required human oversight or iterative prompting. For engineering teams that rely on CI/CD, automation, and parallel agent management, these improvements translate into fewer interruptions and faster iteration cycles.

Real-world partner feedback and benchmarks

Anthropic reports wide-ranging partner feedback and benchmark gains. On a 93-task coding benchmark, Opus 4.7 improved resolution by about 13% over Opus 4.6 and solved several tasks neither Opus 4.6 nor Sonnet 4.6 could complete. Partners such as Replit, Warp, and Quantium reported better code quality, more efficient debugging, and meaningful lifts in production task resolution. Specialized benchmarks and customer use-cases—ranging from document reasoning to enterprise code review—showed double-digit gains in many contexts. Examples called out include stronger recall in code review, fewer tool failures in orchestrated agent workflows, and marked improvements on long-horizon agentic tasks.

Expanded multimodal capability and memory

Opus 4.7 raises the practical bar for vision-enabled workflows. The model accepts images up to 2,576 pixels on the long edge (around 3.75 megapixels), allowing it to handle dense screenshots, complex diagrams, and fine-grained visual references more reliably than prior Claude models. Anthropic also notes better file-system memory for sustained multi-session work: Opus 4.7 can recall important notes across sessions and use them to reduce the need for repeated context, which helps in long-term investigations and ongoing projects.

Safety posture and staged release strategy

Anthropic is explicit that Opus 4.7 was released under a conservative rollout plan tied to Project Glasswing and their Mythos-class research. During training, they experimented with measures to differentially reduce high-risk cyber capabilities in Opus 4.7 relative to Mythos Preview. The model ships with safeguards that automatically detect and block requests indicating prohibited or high-risk cybersecurity uses. Security professionals with legitimate needs—such as vulnerability research, penetration testing, and red-teaming—are invited to join Anthropic’s Cyber Verification Program to get vetted access. Anthropic’s internal safety evaluations show a safety profile similar to Opus 4.6: modest overall improvement on some measures, and areas where behavior can still be improved. They continue to regard Mythos Preview as their most tightly aligned model.

Practical performance highlights and use cases

Anthropic describes a range of concrete wins reported by customers and partners:

Faster median latency and stricter instruction following that help keep developers in the flow for multi-step tasks.
Better handling of ambiguous or incomplete data—reporting missing inputs rather than inventing plausible but incorrect fallbacks.
Significant improvements on document reasoning and law-related benchmarks, including better handling of contract provisions and review tables.
Enhanced agent-team coordination and role fidelity in production orchestrations, reducing the need for step-by-step guidance.
Visual-acuity gains enabling new classes of work, for example, autonomous penetration-testing workflows that were previously impractical.

Availability, pricing, and integrations

Claude Opus 4.7 is available across Anthropic’s products and API, and through major cloud partners including Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry. Anthropic has kept pricing consistent with Opus 4.6: $5 per million input tokens and $25 per million output tokens. Developers can access the model via the Claude API, and Anthropic invites security-focused organizations to apply for the Cyber Verification Program where appropriate.

What this means for teams and next steps

For engineering organizations and product teams, Opus 4.7 represents a practical upgrade path: better code quality, fewer tool errors, and improved long-context performance can reduce manual oversight and accelerate delivery. At the same time, Anthropic’s staged release and safeguards reflect an ongoing trade-off between capability and controlled deployment, particularly for cybersecurity-related tasks. Teams planning to integrate Opus 4.7 should re-evaluate prompt designs—because the model follows instructions more literally—and consider participating in Anthropic’s verification programs if their work touches sensitive domains.

Conclusion

Claude Opus 4.7 is a measured but substantial step up from Opus 4.6, focused on sustained reasoning, engineering workflows, and higher-resolution multimodal understanding. Anthropic’s combination of partner-driven benchmarks, real-world use cases, and cautious safety measures makes this release notable for enterprises that want more dependable agentic behavior without immediately moving to the broadest, most capable Mythos-class models. For many production teams, Opus 4.7 may provide immediate productivity gains while Anthropic continues to refine alignment on higher-capability systems.