5-Min Brief: Anthropic Just Released a New AI Model. Here's What's Different — and What's Still Being Held Back.

What you need to know — in 30 seconds

• Anthropic released a new AI model today called Claude Opus 4.7 — an upgrade from the previous Opus 4.6
• The headline improvements are in coding, vision, and handling long complicated tasks
• It's the first Claude model that can see images in significantly higher resolution
• But the more interesting story is what Anthropic is still holding back — and why

Quick refresher on how Anthropic names things, because it matters for this story. Anthropic makes a family of AI models under the Claude brand. They come in tiers — Haiku is the fast and lightweight one, Sonnet is the middle ground, and Opus is the most powerful one available to the public. Above all of those sits something called Mythos — a model Anthropic has described as too powerful to release broadly. We covered the Mythos story here when Anthropic revealed it had found thousands of security holes in major software systems that humans had missed for decades.

Today's release — Opus 4.7 — sits just below Mythos. Think of it as the most powerful AI Anthropic is willing to let anyone use right now.

What's actually new in Opus 4.7

The biggest improvement most users will notice is in coding — specifically in handling long, complicated coding tasks that previously required a lot of hand-holding.

Early testers describe being able to hand off their hardest coding work to Opus 4.7 and walk away with confidence. One company that tested it early said it lifted their coding task success rate by 13% compared to Opus 4.6 — including solving tasks that neither Opus 4.6 nor Sonnet 4.6 could handle at all.

For non-coders, the more relevant improvement is vision. Opus 4.7 is the first Claude model with high-resolution image support — it can now process images at more than three times the resolution of the previous version. In plain English: it can read charts, diagrams, documents, and images with significantly more detail and accuracy than before.

It's also better at professional knowledge work — document creation, financial analysis, multi-step research — and handles longer tasks more reliably without losing track of where it is.

Pricing is unchanged from the previous version.

The cybersecurity piece

Here's where it gets interesting.

Anthropic has been sitting on Mythos — their most powerful model — partly because it's so capable at finding and exploiting software vulnerabilities that they're not comfortable releasing it broadly. Instead, they've been sharing it with a small group of companies and government agencies to help patch the security holes it finds, before anyone with bad intentions can use those same capabilities.

Opus 4.7 is their first attempt at threading a needle: releasing a genuinely powerful model while building in guardrails specifically around the cybersecurity capabilities that make Mythos so dangerous.

The new safeguards automatically detect and block requests that appear to be trying to use the model for prohibited security exploits. Security professionals with legitimate needs — penetration testers, vulnerability researchers — can apply for a verified program that gives them expanded access.

Anthropic is treating Opus 4.7 as a test bed. What they learn from how these safeguards hold up in the real world will inform whether and when they eventually release Mythos more broadly.

How does it compare to the competition?

The short answer: it's the strongest generally available AI model right now, but the margins are thin.

Opus 4.7 leads GPT-5.4 from OpenAI and Gemini 3.1 Pro from Google on most of the key benchmarks for coding and complex reasoning. But "leads" is doing a lot of work in that sentence — in some categories the gap is only a few percentage points. GPT-5.4 still beats it on certain tasks, particularly web search and some language benchmarks.

This is the reality of frontier AI in 2026: the top models are genuinely very close to each other, and the differences that matter most are often specific to what you're using them for.

What this means for you

If you use Claude through claude.ai on a Pro, Max, Team, or Enterprise plan, Opus 4.7 is available to you today. You don't need to do anything — just select it from the model picker.

If you don't use Claude and are just watching: this is what the normal cadence of AI model releases looks like. Every few months, the major labs push out a new version that's meaningfully better than the last one. Each release raises the floor for what "good AI" means. And the gap between what they're releasing publicly and what they're capable of building — illustrated by Mythos sitting unreleased above Opus 4.7 — keeps getting more interesting.

HumanReadable-AI covers AI news in plain English every weekday. Subscribe below — free, no jargon, always under five minutes.