What does Opus 4.7 really change?
Anthropic released Claude Opus 4.7 on 16 April 2026. It is their new flagship model, the one you use when you open Claude on the paid plan. The main difference from Opus 4.6: it is markedly better at spotting its own reasoning errors before acting. When it plans a complex task, it flushes out its own inconsistencies and corrects them itself, before answering you. The result: fewer corrections for you to make on your side.
Is it really better than the previous version?
The figures published by Anthropic speak for themselves. On a test of 93 programming tasks, Opus 4.7 succeeds on 13% more than Opus 4.6. On real-world code benchmarks, some partners report improvements of up to 70% success against 58% for its predecessor.
There is also a significant visual bonus: Opus 4.7 can analyse images three times more precisely than before, up to 2,576 pixels wide. For AI agents that "look" at your screen in order to click on it (such as Cowrok, OpenClaw), this is a major change.
Anthropic said it clearly in their announcement: Opus 4.7 is less powerful than Claude Mythos Preview, their ultra-secret model still under lock and key. They are using Opus 4.7 as a testing ground for their new safety guardrails. If it holds up in real-world conditions, these protections will be enabled on Mythos to, one day, make it public.
The three technical new features to remember
Beyond the model itself, Anthropic is launching three new features:
- The "xhigh" level: Until now, you could set Claude's thinking effort between "high" and "max". A new intermediate tier is arriving — "extra high" — for a better balance between speed and depth. It is now the default setting in Claude Code.
- The /ultrareview command: In Claude Code, this new command launches a session dedicated to reviewing your code to flush out bugs. Pro and Max subscribers get 3 free trial uses.
- "Task Budgets": You can now set a token limit for Claude on long tasks. It adapts its depth of reasoning so as not to exceed your budget. Useful for keeping costs under control.
Claude Design, what exactly is it?
The next day, 17 April, Anthropic launched Claude Design, a visual creation tool built directly into Claude. The principle: you describe what you want to create (a web page, a presentation, an app mockup), and Claude generates an interactive visual result that you can refine through conversation.
What sets Claude Design apart is that it learns your visual identity from the very first use. It reads your website and your existing files to retain your colours, fonts and components. Everything it creates afterwards automatically follows your brand guidelines.
Once you are happy with the result, you can export it as PDF, PowerPoint, HTML, or send it directly to Claude Code so that it turns it into real production code.
For founders preparing a pitch deck, product managers who want to turn an idea into a mockup without waiting for a designer, and marketers creating campaign visuals. Professional designers use it to quickly explore more creative directions before choosing.
How much does it cost?
Good news: Opus 4.7 is the same price as Opus 4.6. No increase. Claude Design is included in paid subscriptions, at no extra cost.
| Plan | Price | Claude Design included? |
|---|---|---|
| Free | €0 | ❌ No |
| Pro | ~$20/month | ✅ Yes |
| Max / Team / Enterprise | $100–200/month | ✅ Yes + extended usage available |
For developers using the API directly: $5 per million input tokens, $25 for output, identical to Opus 4.6.