Claude Code Adds Opus 4.7 + /ultrareview for Better Agentic Coding
Claude Code's v2.1.107-111 update integrates Opus 4.7 (10-15% higher task success, xhigh effort tier), /ultrareview (parallel multi-agent reviews, 3 free for Pro/Max), 1-hour prompt cache TTL, and UI fixes—run `claude update` to cut token costs and boost long-horizon reasoning.
Opus 4.7 Improves Reasoning and Vision for Agentic Tasks
Claude Opus 4.7, Anthropic's new flagship coding model, now ships natively in Claude Code, delivering 10-15% higher task success rates and fewer tool errors per early testers. It excels in benchmarks like Finance Agent (state-of-the-art), GDPval-AA (top results), coding, vision, document reasoning, and long-context tasks. Vision capacity triples to ~3.75 megapixels, enabling better computer-use agents and data extraction from large images. Use the new xhigh effort tier—accessed via model picker, /effort xhigh, or --effort—for balanced reasoning depth and speed in long agentic coding runs, positioned between high and max to avoid max latency costs. Pricing holds at $5/million input tokens and $25/million output; token usage rises modestly (1.0-1.35x) due to tokenizer updates but improves efficiency at similar effort levels. File system memory enhancements support multi-session work, and auto mode extends to max users for longer autonomous tasks.
/ultrareview Enables Parallel Multi-Agent Code Reviews
/ultrareview runs parallel multi-agent reviews across diffs to catch non-compiler bugs; Pro/Max users get three free reviews per billing cycle. Pair it with /less-permission-prompts, which scans transcripts to auto-propose tool allowlists, reducing repetitive approvals. /recap provides session context on resume (configurable via /config). The skill tool now auto-discovers and invokes built-in slash commands like init, statusline, review, security, insights, and onboarding for automated execution. Custom rules override built-in defaults, letting you tailor behavior precisely.
UI and Performance Fixes Boost Reliability
/tui fullscreen delivers flicker-free terminal rendering with autoScrollEnabled config, fixing iTerm2 + tmux tearing and improving /context grid display—/clear no longer drops session names. Set ENABLE_PROMPT_CACHING_1H for 1-hour TTL across API key, Bedrock, Vertex, and Foundry, slashing input costs for hour-plus sessions. Other wins: @ file suggestions avoid full project rescans, LSP diagnostics show post-edits, MCP avoids connection-drop hangs, and extended thinking indicators add rotating progress hints. Update via claude update; the open-source repo has 75K+ stars.