Anthropic Releases Claude Opus 4.7: Improved Coding, Coordination, and Security

Anthropic has released a new version of its Claude language model, Opus 4.7, with significant improvements in coding capabilities, multi-agent coordination, and cybersecurity safeguards.

Details

Anthropic's new Claude Opus 4.7 model brings significant improvements in coding capabilities, with a 6.8-point jump in the SWE-bench Verified benchmark and a 13% lift in Hex's internal 93-task coding suite. The model also introduces multi-agent coordination, allowing it to orchestrate parallel AI workstreams instead of just chaining tasks sequentially. This is a meaningful architectural shift for complex automation. Additionally, Opus 4.7 has better vision capabilities, with support for images up to 3.75 megapixels, and improved file-system memory for maintaining state across sessions. The model also claims to be the 'most literally instruction-following Claude model ever', with less creative interpretation of prompts. Importantly, Opus 4.7 ships with automatic detection and blocking of prohibited cybersecurity uses, addressing concerns raised in the recent 'Claude Mythos' story. While the pricing remains the same as the previous version, the updated tokenizer can lead to a 35% hidden cost increase for high-volume users, so they should run their own tokenizer comparisons before migrating.