Claude Opus 4.7 Is Here: Better at Hard Coding, Vision, and Actually Catching Its Own Mistakes

Anthropic just dropped Claude Opus 4.7, and it’s a solid step up from Opus 4.6—especially if you’re doing the kind of coding work that usually requires a senior dev breathing down the model’s neck.

The headline improvements are in advanced software engineering, vision, and multi-step task handling. But what caught my attention is how testers describe it: less hand-holding, more self-correction. One tester from a financial platform said the model catches its own logical faults during planning. That’s the kind of thing that actually saves time, not just sounds good in a press release.

Vision got a real upgrade too. Higher resolution support means it can actually read chemical structures, complex diagrams, and technical docs. Solve Intelligence, a company working on life sciences patents, called this out specifically. If you’ve ever tried to get a model to parse a messy flowchart or a handwritten note, you know how big a deal this is.

There’s also a notable improvement in “taste” for professional outputs—interfaces, slides, documents. That’s a subjective metric, but multiple testers mentioned it, so I’m inclined to believe it’s real.

The Cyber Safeguard Angle

Here’s where it gets interesting. Last week Anthropic announced Project Glasswing, which is all about AI risks in cybersecurity. Opus 4.7 is the first model shipping with safeguards that automatically detect and block high-risk cybersecurity requests. The idea is to test these protections on a less capable model before rolling them out to their top-tier model, Claude Mythos Preview.

Security professionals who need legitimate access (vulnerability research, pen testing, red-teaming) can join Anthropic’s new Cyber Verification Program. It’s a sensible approach—let the good guys in, keep the script kiddies out.

Pricing and Availability

Pricing hasn’t changed: $5 per million input tokens, $25 per million output tokens. Same as Opus 4.6. Available across all Claude products, API, Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry. The API model name is claude-opus-4-7.

What Testers Are Saying

The early access feedback is unusually specific and consistent. Here’s what stood out:

• Hex called it the strongest model they’ve evaluated, noting it correctly reports missing data instead of making stuff up—a trap Opus 4.6 still falls into.
• Devin said it takes long-horizon autonomy to a new level, working coherently for hours and pushing through hard problems instead of giving up.
• Replit called it an easy upgrade decision.
• On a 93-task coding benchmark, Opus 4.7 lifted resolution by 13% over Opus 4.6, including four tasks neither Opus 4.6 nor Sonnet 4.6 could solve.
• On a research-agent benchmark, it tied for top overall score at 0.715 and showed the most consistent long-context performance of any model tested.

One tester summarized it well: “low-effort Opus 4.7 is roughly equivalent to medium-effort Opus 4.6.” That’s the kind of efficiency gain that actually matters in day-to-day work.

The Bottom Line

Opus 4.7 isn’t a revolutionary leap, but it’s a meaningful one. Better coding, better vision, better at long-running tasks, and it comes with real-world cybersecurity safeguards that Anthropic can iterate on. If you’re already using Opus 4.6, this is a no-brainer upgrade. If you’ve been on the fence about Claude for hard coding work, this might be the version that changes your mind.