Anthropic launches Claude Opus 4.7 with a 13% improvement in coding

Anthropic unveiled Claude Opus 4.7 — the next generation of its flagship model. The release occurred in April 2026. This is not a revolution, but an improvement in the areas that truly pain developers and companies.

Coding at a New Level

On a test benchmark of 93 programming tasks, Opus 4.7 scored 13% higher than Opus 4.6. More specifically: it solved four tasks from the set that are beyond even the faster Sonnet 4.6.

But the numbers matter least. More important are the reviews from developers who tested it. They say: now you can take the most convoluted, multi-step task — one that previously required human breakdown — and simply give it to Opus 4.7. The model grasps context, self-verifies, requires no oversight or rework.

This is particularly evident in long asynchronous workflows: CI/CD pipelines, various automations, microservices. Opus 4.7 thinks deeper than previous models, offers its own perspective rather than simply nodding along with user ideas.

According to one tester, low-effort Opus 4.7 is roughly equivalent to medium-effort Opus 4.6.

Vision and Taste

Image vision has improved: the model now sees at higher resolution. This helps in analyzing complex diagrams, system screenshots, architectural drawings, where every line matters. And even more interesting: the model has become more aesthetically refined in visual tasks. When you ask it to create an interface, presentation, or document — the results look more beautiful and professional on the first try.

This doesn't mean it was ugly before. It's just that now interfaces look like designer work, not a system for generating rough mockups.

• Analysis of complex diagrams in high resolution
• User interface design with better aesthetics
• Slides and documents look more polished and professional
• All of this with no extra charge — pricing remains the same

Cybersecurity Built In

Anthropric took security seriously. The company's Project Glasswing identified risks associated with using large models in cybersecurity. Therefore, Opus 4.7 comes with built-in protections: the system automatically catches and blocks requests for dangerous and prohibited purposes.

But this is not a complete ban. Anthropic understands that legitimate penetration testers, vulnerability researchers, and red teams are needed for defense. There is a Cyber Verification Program for them: you can get better access to the model if you work for security, research, or red testing purposes.

Availability and Pricing

Opus 4.7 is already available everywhere: on Claude.ai, via Claude API, in Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. There are no regional restrictions — the model is available to users worldwide.

Pricing has not changed compared to Opus 4.6: $5 per million input tokens, $25 per million output tokens. If you previously budgeted for 4.6, now you get 4.7 for the same price — a de facto free upgrade.

In code, the model is specified as `claude-opus-4-7`.

Tester Reviews

Feedback from developers was very positive. A fintech team working with millions of users noted that the combination of speed and accuracy could be a game-changer. The Hex team (an analytics platform) praised the model for correctly reporting when data is insufficient, rather than making up answers. This is especially valuable for systems where errors are costly.

How Work Changes

The world of code automation is changing before our eyes. Previously, LLMs helped with minor tasks: suggesting variable names, completing functions, explaining complex code. Now you can delegate entire workflows — tasks that took days now complete in hours. This is not a replacement for developers. It's a shift in how they spend their time. Before: tedious, repetitive. Now: architecture, problem-solving, team leadership. The model handles mechanics; humans focus on strategy.