Anthropic has launched Claude Opus 4.7, its latest AI model, focusing on enhanced safety features and improved performance across various applications. While offering significant advancements in areas like software engineering and instruction following, the model incorporates new safeguards designed to detect and block high-risk cybersecurity uses, marking a step towards responsible AI deployment.
Key Takeaways
- Claude Opus 4.7 boasts improved software engineering, instruction fidelity, and multimodal processing capabilities.
- New automated safeguards are integrated to detect and block prohibited or high-risk cybersecurity uses.
- The model is less broadly capable in cybersecurity than the experimental Claude Mythos Preview.
- Anthropic is using Opus 4.7 to test safeguards for future, more powerful models.
- The model is now generally available across Anthropic’s products and major cloud platforms.
Enhanced Capabilities and Safety Measures
Claude Opus 4.7 represents a notable upgrade from its predecessor, Opus 4.6. The new model demonstrates superior performance in advanced software engineering, handling complex, long-running tasks with greater consistency and precision. It also shows significant improvements in multimodal processing, capable of accepting images up to 2,576 pixels on the long edge, which is more than three times the resolution of previous Claude models. This enhanced vision capability supports more detailed data extraction and analysis from complex visual inputs.
Anthropic has also refined the model’s instruction-following behaviour, making it more literal and precise. This requires users to re-tune their prompts and harnesses when migrating from older versions. Furthermore, Opus 4.7 exhibits improved memory management, remembering important notes across long, multi-session work, thereby reducing the need for extensive upfront context.
A key focus of this release is the integration of new cybersecurity safeguards. Anthropic has intentionally "differentially reduced" the cyber capabilities of Opus 4.7 compared to its experimental Claude Mythos Preview. The model is equipped with automated systems that detect and block requests indicating prohibited or high-risk cybersecurity uses. This approach allows Anthropic to learn from real-world deployments of these safeguards, informing their strategy for the eventual broader release of more powerful Mythos-class models.
Availability and Pricing
Claude Opus 4.7 is now generally available through all of Anthropic’s Claude products, its application programming interface (API), and via cloud providers including Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry. The pricing remains consistent with Claude Opus 4.6, set at $5 per million input tokens and $25 per million output tokens. Anthropic is also encouraging security professionals interested in using the model for legitimate cybersecurity purposes, such as vulnerability research and penetration testing, to apply through a formal verification program.
Safety Profile and Future Outlook
Anthropic continues to prioritise safety and responsible AI development. While Opus 4.7 shows a similar safety profile to Opus 4.6, with low rates of concerning behaviour, it demonstrates improvements in honesty and resistance to malicious prompt injection attacks. The company’s alignment assessment indicates that the model is "largely well-aligned and trustworthy." The insights gained from Opus 4.7’s deployment will be crucial for Anthropic’s long-term goal of safely scaling advanced AI models.
