Anthropic has officially released Claude Opus 4.7, opening access immediately and positioning the new model around stronger software engineering performance. The update focuses on handling difficult coding work, long-running tasks, and more capable multi-step reasoning compared with the earlier Opus 4.6.
Software engineering performance gets a major upgrade
According to Anthropic, Opus 4.7 shows clear gains on advanced software engineering tasks. The model is designed to handle complex and long-duration asynchronous workloads more reliably, including automation workflows and CI/CD jobs. Anthropic also said the system can catch its own logic issues during planning and verify code outputs before reporting results. Early feedback from technology leaders at Hex, Notion, and Replit described the model as highly effective for both execution and deeper technical discussion.
Higher-resolution image input expands visual analysis
Anthropic also upgraded the model’s multimodal vision capabilities. Opus 4.7 can now process images up to 2,576 pixels on the long side, or roughly 3.75 megapixels, which the company said is more than three times the capacity of previous Claude models. That increase is aimed at tasks requiring fine visual detail, such as reading dense software screenshots, technical charts, and chemical structures.
Pricing stays the same as Opus 4.6
Despite the capability increase, Anthropic kept API pricing unchanged at $5 per million input tokens and $25 per million output tokens. Developers can access the model through Claude products, Anthropic’s API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
The company also introduced new cybersecurity guardrails that automatically detect and block high-risk or policy-violating cyberattack requests. Security professionals with legitimate use cases, including vulnerability research and penetration testing, are being directed to the Cyber Verification Program. Opus 4.7 also adds a new xhigh reasoning level, placed between high and max. Anthropic noted that the updated tokenizer may produce slightly higher token counts for the same text input.