Claude Code: Opus 4.7 + /ultra Review Boost Coding

Opus 4.7 Balances Speed, Depth, and Capabilities

Claude Opus 4.7 introduces /effort XI tier as a middle option between high and max efforts, ideal for long-horizon coding without full max latency costs—access via model picker or /effort XI flag. Vision support triples to 3.75 megapixels for handling larger images. File system memory improves for multi-session work, auto mode extends to max users for longer autonomous tasks, and benchmarks show across-the-board gains in finance agent, GPQA, coding, vision, and long-context reasoning. Early tests report 10-15% higher task success rates and fewer tool errors versus 4.6, with pricing unchanged at $5/million input and $25/million output tokens; expect 1-1.35x higher usage from tokenizer updates.

Parallel /ultra Review Catches Hidden Bugs

/ultra review runs multiple agents in parallel across code diffs to detect compiler-missed bugs, with Pro/Max users getting 3 free reviews. Pair it with custom rules that override built-in defaults for tailored checks. Skill tool now discovers and invokes all built-in /commands automatically, enabling real automated execution.

Workflow Polish: Commands, Caching, and Fixes

Slash less permission scans transcripts to build reusable allow lists, cutting repetitive approvals. /recap provides session context on resume, /TUI full screen eliminates flicker for terminal rendering with auto-theme syncing to dark/light modes (iTerm2/marks fixed). Prompt cache TTL extends to 1 hour via CACHING_ONE_H environment variable across API key, Bedrock, Vertex, and Foundry providers. Fixes include add-file suggestions skipping full project rescans, correct LSP diagnostics post-edits, no hangs on MCP tool drops, cleaner /context grid, and /clear preserving session names. Run claude update to apply all 10 features and 45+ fixes.