Anthropic Launches Claude Opus 4.7 - Self-Verification, 3.3x Higher-Resolution Vision, 13% Coding Improvement

Anthropic has unveiled its next-generation flagship model, Claude Opus 4.7.

Key Performance and Benchmarks

• It recorded 87.6% on SWE-bench Verified, ranking #1 ahead of GPT-5.4 (57.7%).
• On coding benchmarks, it showed a 13% improvement over the previous version, Opus 4.6, with production task resolution rates increasing 3x.
• It achieved strong results on key metrics including GPQA Diamond 94.2%, Terminal-Bench 2.0 69.4%, and Finance Agent 64.4%.

Key New Features and Improvements

• Self-Verification: A feature was added that allows the model to design and execute its own verification steps before completing a task.
• High-Resolution Vision: It supports 3.3x higher-resolution vision up to 2,576px, with a new tokenizer applied.
• Agentic Reasoning: With the addition of the 'xhigh effort' level, agentic reasoning ability improved by 14%, and tool-use errors decreased to 1/3.

Availability Pricing is the same as Opus 4.6 at $5/$25 per MTok, and it is available through Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.