Every New Opus 4.7 Feature, Explained

Claude Opus 4.7 shipped with 12 new features worth knowing about, and most people only heard about 5 of them. Here's every one explained in plain English, no benchmark jargon, with how to turn each on.

What You'll Learn

• Auto Mode, /ultrareview, /fewer-permission-prompts, Recaps, /focus and /effort
• 3x better vision, improved design output, memory across sessions
• Task budgets, self-verification, better instruction following
• How to use /effort to lower your token usage on simple tasks

1. Auto Mode

Claude now figures out which actions are safe during long running tasks and just does them without asking you. Before this, Claude stopped and asked permission for every single action on your computer. Now you hand it a task, walk away, and come back to finished work. Available on Max, Team, and Enterprise plans. A safety classifier reviews each action before it runs and blocks anything risky.

2. /ultrareview

New slash command. Type /ultrareview and Claude sends your code to 4 AI agents at once. One checks security, one checks code logic, one checks speed, and one checks design style. You get one report back with everything they found. Like having 4 expert reviewers on call. Pro and Max users get 3 free ultrareviews to try it out.

3. /fewer-permission-prompts

Type this command and Claude scans your history to find all the permissions it kept asking you about. Then it builds a list of your most approved permissions and adds them to an auto-allow list so it stops interrupting you. One-time setup. After that, the actions you always approve just happen automatically.

4. Recaps

Start a long task, step away, come back, and Claude gives you a short summary of what it did and what it's working on next. Way more token-efficient than scrolling through the whole conversation to figure out where things are at. You can also type /recap anytime to get caught up. Configurable in /config or with the CLAUDE_CODE_ENABLE_AWAY_SUMMARY environment variable.

5. /focus And /effort

/focus tells Claude to hide all the in-between work it does (tool calls, file reads, intermediate steps) so you only see the final result. Clean output.

/effort lets you quickly control how hard Claude thinks. Use xhigh for complex tasks that need deep reasoning. Use lower settings for quick answers and to save on token usage. Anthropic raised the default effort to xhigh across all Claude Code plans.

6. 3x Better Vision

Maximum image resolution went from 1568 pixels to 2576 pixels. That's over 3x the detail. Screenshots, designs, diagrams, documents: Claude can now see fine details that were blurry before. Computer use and screen reading actually work properly. Coordinates are now 1:1 with actual pixels so Claude knows exactly where things are on your screen.

7. Better Design Output

Interfaces, slides, presentations, documents. Opus 4.7 produces output that looks professionally designed instead of generic AI output. Lovable called it "the best model in the world for building dashboards and data-rich interfaces" and said the design taste is genuinely surprising. If you're building landing pages, dashboards, or slide decks with Claude, this is a noticeable upgrade.

8. Memory Across Sessions

Opus 4.7 is better at writing and reading its own notes between sessions. If Claude keeps a memory file or scratchpad, it actually uses that information to pick up where it left off. Less re-explaining at the start of every session, less context you have to provide manually. Works especially well with CLAUDE.md files and the built-in memory system.

9. Task Budgets

New feature for developers using the API. You can now give Claude a token budget for an entire task. Claude sees a running countdown and uses it to prioritize work and finish gracefully as the budget runs out instead of cutting off mid-task. Think of it like giving an employee a time limit: they scope the work to fit instead of running over. Useful for controlling costs on long agentic runs.

10. Self-Verification

Opus 4.7 checks its own work before sending it to you. It goes back through what it built, looks for mistakes, and fixes them before reporting back. Vercel said it "even does proofs on systems code before starting work, which is new behavior we haven't seen from earlier Claude models." This is the feature that makes it safe to hand Claude hard tasks and walk away.

11. Better Instruction Following

Opus 4.7 takes your instructions literally. Previous models would interpret loosely or skip parts entirely. Now it follows exactly what you say. One thing to note: prompts written for older models might need updating because Claude no longer fills in gaps or guesses what you meant. Be specific and you get exactly what you asked for.

12. Cyber Verification Program

Opus 4.7 has built-in safeguards that detect and block high-risk cybersecurity requests. If you're a security professional who needs to use Claude for legitimate penetration testing, vulnerability research, or red-teaming, you can apply for the new Cyber Verification Program at claude.com/form/cyber-use-case. This is Anthropic's first model with active cyber safeguards, and what they learn here will help them release their more powerful Mythos model later.

Go Deeper

Printable PDF version of this guide: https://drive.google.com/file/d/1K5X88p1SMGudw_jIwEN5c1nV8RVP6MXY/view?usp=drivesdk

Want the whole system installed live, with Cooper on the call? Build Your Own AI, the two week sprint runs the first two weeks of August.

Or get a new working tool every Monday inside The AI Operator community on Whop.