Claude Opus 4.7: 87.6% on SWE-Bench, new “xhigh” level and Claude Design

Claude Opus 4.7: 87.6% on SWE-Bench, new “xhigh” level and Claude Design

N Equipo NodoAI
4 min read

Anthropic has launched Claude Opus 4.7, its top model, with a significant leap in code: 87.6% on SWE-Bench Verified (13 points above Opus 4.6), a new level of reasoning called xhigh, and a new product called Claude Design that generates design and prototypes. All at the same price as the previous version.

I’ll tell you about it without marketing: what it does, what changes for you, and where it still lags behind the competition.

What’s big about this release

Three things with real impact on how you work with Claude:

  • More reliable coding. 87.6% on SWE-Bench Verified, the most serious benchmark for “does this model fix real bugs?”. Sustained improvement in long and complex software tasks.
  • Xhigh level. Between high and max there’s a new step. It gives you fine control between speed and depth of reasoning without having to go to the maximum cost.
  • 3× higher resolution vision. Same API, same price, but the model “sees” images with much more detail. Important if you work with captures, diagrams, or scanned documents.

Claude Design: the surprise product

Anthropic has taken advantage of the launch to introduce Claude Design, a separate product within Anthropic Labs that lets you “design by conversing”: slides, one-pagers, prototypes, visual mockups. It’s not Figma or Canva — it’s something closer to “explain what you want and I’ll set it up for you”.

It’s in lab state (early access), so the final quality is yet to be seen. But the direction is clear: Anthropic wants to get out of the “API for devs” niche and into visual products.

Claude Code: redesigned desktop app

If you used Claude Code in the terminal, the new desktop app will appeal to you:

  • Parallel tasks with a session sidebar (multiple projects at once).
  • Drag & drop workspace: you drag folders and files directly into the conversation.
  • Integrated file editor with faster diffs.
  • Native SSH support on Mac: you connect to remote servers without leaving the client.

For developers who already do pair programming with Claude, this is noticeable from day one.

Claude Security (beta · Enterprise)

Another important novelty: Anthropic has released Claude Security in beta for Enterprise clients. It scans code for vulnerabilities and proposes fixes with Opus 4.7. It’s in line with what GitHub Advanced Security or Snyk do, but with LLM generating the patches.

Pricing: no changes

This is the strangest and most welcome thing: Opus 4.7 maintains the price of Opus 4.6.

  • $5/M input tokens
  • $25/M output tokens

It’s a better model for the same money, which is exceptional in this market.

What’s still lagging behind

Anthropic itself has acknowledged in its communication that Opus 4.7 is still below a model not yet launched internally, dubbed Mythos. In general benchmarks, GPT-5.5 still scores slightly higher, although the differences are statistical noise in many cases.

Quick comparison: Opus 4.7 vs GPT-5.5

Criteria Claude Opus 4.7 GPT-5.5
Coding (SWE-Bench) 87.6% ✅ ~84%
Deep reasoning Better (xhigh) Thinking mode
Response speed Average High ✅
Product ecosystem Code, Design, Security Wider ✅
Input/Output price $5/$25 per M Similar

Full analysis in our ChatGPT vs Claude · 2026 comparison.

FAQ

How do I access Claude Opus 4.7?

Claude Pro plan (20$/month) grants access. It’s also available in the API under the name claude-opus-4-7.

What does the “xhigh” level mean?

It’s a new step between high and max reasoning effort. Deeper than high, less slow and expensive than max.

Is Claude Design already available?

Yes, but in early access within Anthropic Labs. Functionality and stability may change.

Is it worth upgrading from 4.6 to 4.7 if I already pay for Pro?

Yes, at no extra cost. Better coding, better vision, same price. There’s no reason not to use it.

Does it replace ChatGPT?

It depends on the case. If you do a lot of coding or long texts: Claude probably. If you want a wider ecosystem (image, voice, Sora): ChatGPT. Details in the full comparison.

Next step

If you’re starting with LLMs: free beginner course. If you already handle Claude and want to dive deeper into coding and agents, the advanced course covers APIs, Cursor, and product building. And if you want to work with Claude professionally: IA Developer guide.

Sources: Anthropic · Claude Opus 4.7 · GitHub Changelog.

N
Equipo NodoAI
Equipo editorial · NodoAI

Equipo editorial de NodoAI. Especialistas en inteligencia artificial, automatización y productividad para profesionales hispanohablantes.

Recibe más contenido como este en tu inbox.

Sin spam. Sin hype. Solo lo que importa en IA.