Claude Opus 4.7 Is Here: Better at Hard Coding, Vision, and Not Making Dumb Mistakes

Anthropic just dropped Claude Opus 4.7, and it’s now generally available across all their products and major cloud platforms. The pricing stays the same as Opus 4.6 — $5 per million input tokens and $25 per million output tokens — which is nice given the upgrades.

So what’s actually better? The headline improvements are in advanced software engineering, particularly for the really gnarly tasks that developers previously had to babysit. Early testers are saying they can now hand off their hardest coding work with actual confidence. The model is more rigorous with long-running tasks, pays closer attention to instructions, and — this is the interesting part — it can verify its own outputs before reporting back. That’s a big deal for anyone who’s tired of reviewing hallucinated code.

Vision also got a substantial bump. The model can see images in higher resolution now, and it’s producing better quality interfaces, slides, and docs. It’s more tasteful and creative, which sounds subjective but apparently shows up in the work product.

One thing worth noting: Opus 4.7 is

t less broadly capable than Anthropic’s most powerful model, Claude Mythos Preview. That model is still on limited release due to cybersecurity concerns. Opus 4.7 sits below it in cyber capabilities — Anthropic actually experimented with differentially reducing those capabilities during training. It ships with new safeguards that automatically detect and block prohibited or high-risk cybersecurity requests. Security professionals who need it for legitimate work (vulnerability research, penetration testing, red-teaming) can apply for Anthropic’s new Cyber Verification Program.

What Early Testers Are Saying

The early-access feedback is unusually detailed and positive. Here’s the gist:

A financial tech platform noted that Opus 4.7 catches its own logical faults during planning and accelerates execution, which is exactly what you want when you’re serving millions of consumers. Another tester said it stands out not just for raw capability but for handling real-world async workflows — automations, CI/CD, long-running tasks — and that it brings a more opinionated perspective rather than just agreeing with the user.

Hex, the data platform, called it the strongest model they’ve evaluated. The key improvement? It correctly reports when data is missing instead of providing plausible-but-incorrect fallbacks, and it resists dissonant-data traps that even Opus 4.6 falls for. That’s a genuinely important behavioral improvement that benchmarks don’t always capture.

On a 93-task coding benchmark, Opus 4.7 lifted resolution by 13% over Opus 4.6, including four tasks that neither Opus 4.6 nor Sonnet 4.6 could solve. Combined with faster median latency and stricter instruction following, it’s particularly meaningful for complex, multi-step coding workflows.

Another tester reported the strongest efficiency baseline they’d seen for multi-step work, with consistent long-context performance. On their General Finance module, it scored 0.813 versus Opus 4.6’s 0.767. And on deductive logic — an area where Opus 4.6 struggled — it’s now solid.

Devin‘s team said Opus 4.7 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.

The vision improvements are also getting attention: one tester reported major gains in multimodal understanding, from reading chemical structures to interpreting complex technical diagrams.

Should You Upgrade?

If you’re already on Opus 4.6 and doing serious coding work, this looks like a no-brainer upgrade. The improvements in self-verification, long-context consistency, and handling of complex multi-step tasks are the kind of practical gains that actually reduce frustration. The pricing hasn’t changed, so you’re getting more capability for the same cost.

The cyber safeguards might annoy some power users, but the Cyber Verification Program seems like a reasonable middle ground. And honestly, I’d rather have a model that’s been deliberately tuned to be less dangerous out of the box than one that wasn’t.

Opus 4.7 is available now on all Claude products, the API, Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry. Developers can hit claude-opus-4-7 via the API.