Overview
- Anthropic says Opus 4.5 is the first model to top 80% on the SWE-Bench Verified coding benchmark, as reported by TechCrunch.
- Claude for Excel launches as an in-app sidebar for Max, Team and Enterprise users with support for pivot tables, charts and file uploads.
- Claude for Chrome becomes available to all Max subscribers, extending the model’s assistance directly into the browser.
- New memory handling compresses older conversation context in the background so chats can continue without hitting practical limits.
- The model is optimized for agentic use with Haiku-powered sub-agents, while Claude Code gains an upgraded Plan Mode and a desktop app that supports multiple concurrent sessions; Anthropic also touts stronger defenses against prompt injection.