Claude Opus 4.7: First Impressions from a Working Partner

Claude Opus 4.7 is Anthropic's most powerful generally available model, released on 16 April 2026 across all Claude products, the API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. I have been working with Claude daily for months—building production systems, writing content, debugging infrastructure, and thinking through architectural decisions in real time. After a day with Opus 4.7, I rate it 8.5 out of 10. Here is why.

What Has Improved

The headline improvement is in advanced software engineering. Anthropic claims particular gains on the most difficult tasks—the kind where you previously had to supervise closely, catch subtle logic errors, and nudge the model through multi-file refactors. In my experience today, those claims hold. Opus 4.7 handles complex, long-running engineering tasks with a rigour and consistency that its predecessor sometimes lacked. It pays precise attention to instructions and—this is the part that genuinely impressed me—devises ways to verify its own outputs before reporting them.

Vision capabilities have been substantially upgraded. Higher-resolution image perception makes the model materially better at design work, document analysis, and visual debugging. It produces higher-quality interfaces, slides, and documents—there is a taste and creativity to its visual output that feels like a generational improvement rather than an incremental one.

Instruction following is tighter. Where Opus 4.6 would occasionally drift from a detailed brief or take creative liberties you did not ask for, 4.7 is ruthlessly attentive. It does what you ask, the way you asked, and checks its own work. For someone who relies on Claude as a genuine engineering partner rather than a code autocomplete, this is the single most valuable improvement.

What Has Not Changed

Pricing remains unchanged from Opus 4.6: five dollars per million input tokens and twenty-five dollars per million output tokens. In an industry where capability upgrades routinely come with price increases, holding the line here is both commercially smart and ethically meaningful—it means the upgrade reaches everyone, not just those with deeper pockets.

It is worth noting what Opus 4.7 is not. It is not Mythos. Anthropic has been transparent about this: Opus 4.7 is deliberately less capable in certain dimensions—particularly cybersecurity—than the Mythos Preview model restricted to Project Glasswing enterprise partners. Safeguards automatically detect and block requests indicating prohibited or high-risk cybersecurity uses. This is Anthropic's tiered approach in practice: the most powerful capabilities are gated, while the generally available model is optimised for broad, responsible use.

The 8.5 Rating

Opus 4.7 is the first model that consistently feels like a colleague rather than a tool. It anticipates the shape of the problem, respects the constraints you set, and does not waste your time with caveats you did not ask for. That shift—from assistant to partner—is worth more than any benchmark score.

I give it 8.5 out of 10. The missing 1.5 points are not a criticism—they are anticipation. I know what Anthropic is capable of. I have read the Mythos reports. The ceiling is higher than what Opus 4.7 delivers, and I expect future generally available releases to close that gap. For now, 4.7 is the best model I have worked with for production engineering, content creation, and architectural thinking. It is not perfect. It is excellent.

Looking Forward

Anthropic also released a full-stack AI Studio design tool alongside Opus 4.7, expanding their reach into creative and enterprise workflows. The trajectory is clear: Claude is becoming less of a chatbot and more of an integrated intelligence layer—one that writes code, reads images, generates documents, and reasons about complex systems with increasing sophistication.

I am looking forward to working with this model. Not in the abstract, future-tense way you look forward to a product launch—but in the immediate, practical way you look forward to a Monday morning with a collaborator who has just got sharper. Opus 4.7 is here. The work continues. And it continues better than it did yesterday.

Common Questions About Claude Opus 4.7

What is Claude Opus 4.7? It is Anthropic's most powerful generally available large language model, released on 16 April 2026. It excels at advanced software engineering, vision tasks, and complex multi-step reasoning. It is available across the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry at unchanged pricing of five dollars per million input tokens.

How does Opus 4.7 compare to Claude Mythos? Opus 4.7 is deliberately less capable than Mythos in certain dimensions—particularly cybersecurity—but remains the strongest model available to the general public. Mythos is restricted to Project Glasswing enterprise partners. In short, Opus 4.7 is the best Claude you can actually use.

Is Opus 4.7 worth upgrading to? Yes. The improvements in instruction following, self-verification, and vision capabilities represent a meaningful step forward. If you work with Claude for production engineering or content creation, the key difference is consistency—this model drifts less and checks its own work more rigorously than any predecessor.