Anthropic just released Claude Opus 4.7—and it’s a meaningful leap forward from Opus 4.6. Whether you’re a prompt engineer, a content creator, or someone who relies on AI for daily work, this model deserves your attention.
What’s New in Claude Opus 4.7?
Claude Opus 4.7 is Anthropic’s latest flagship model, now generally available at the same pricing as Opus 4.6. The headline improvements span four key areas:
- Advanced Software Engineering — Opus 4.7 handles complex, long-running coding tasks with minimal supervision. It self-verifies outputs, catches logical faults during planning, and actively pushes back on poor decisions.
- 3x Higher Visual Resolution — Images up to 2,576 pixels on the long edge are now supported, unlocking fine visual detail for computer-use agents, data extraction, and diagram analysis.
- More Creative Professional Deliverables — Higher-quality interfaces, slides, documentation, and dashboards. Anthropic calls it “more tasteful.”
- Stronger Instruction Following — Opus 4.7 takes instructions literally. Prompts written for earlier models may produce unexpected results—time to re-tune your templates.
Benchmark Results That Stand Out
The numbers tell a compelling story:
- CursorBench: 70% vs. 58% (Opus 4.6) — a major jump for AI-assisted coding agents
- Coding benchmark: +13% resolution over Opus 4.6
- Tool error rate: 1/3 of Opus 4.6’s rate
- Visual acuity: 98.5% vs. 54.5% on Opus 4.6
- OfficeQA Pro: 21% fewer errors than Opus 4.6
Early testers echo these results. Harvey hit 90.9% accuracy on BigLaw Bench. Hex summed it up: “Low-effort Opus 4.7 equals medium-effort Opus 4.6.”
A New Effort Level: xhigh
Opus 4.7 introduces a new effort level between high and max: xhigh. Anthropic recommends starting with high or xhigh effort for coding and agentic use cases.
Built-In Cybersecurity Safeguards
Opus 4.7 is the first Anthropic model with automated cyber capability safeguards, automatically detecting and blocking high-risk cybersecurity requests. Legitimate use cases can apply for the Cyber Verification Program.
What This Means for Your Prompts
If you have a library of Claude prompts, review them carefully. Opus 4.7 follows instructions more literally, so outputs may differ from earlier models even with identical prompts. Key updates to consider:
- Be more explicit — Vague instructions lead to literal, sometimes unhelpful responses.
- Re-test agentic workflows — Lower tool error rates and better self-verification mean multi-step prompts may produce significantly better results.
- Leverage the new vision — High-resolution image inputs unlock prompt use cases that were not practical before.
Claude Code Gets Better Too
Opus 4.7 ships alongside new Claude Code features: /ultrareview for dedicated code review sessions, extended Auto mode for Max users, and Claude Code Channels with Telegram and Discord integration.
Migration Note: Tokenizer Update
Opus 4.7 uses an updated tokenizer—the same input maps to roughly 1.0-1.35x more tokens depending on content. At higher effort levels, the model also thinks more. Anthropic recommends measuring the difference on real traffic before fully migrating from Opus 4.6.
Should You Upgrade?
If you use Claude for coding, complex agentic workflows, vision tasks, or high-stakes professional writing—yes. The improvements in self-verification, instruction following, and tool use are substantial. If you want the best Anthropic has to offer, Opus 4.7 is now that model.