At a glance:
Anthropic released Claude Opus 4.7 on April 16, 2026. It leads GPT-5.4 and Gemini 3.1 Pro on most coding and knowledge benchmarks.
Vision jumped to 3.75 megapixels, three times sharper than before. It can now read dense design mockups, diagrams, and screenshots at actual fidelity.
Alongside the model, Anthropic shipped a design tool that generates websites, landing pages, and presentations from text prompts, plus a Figma integration to make the output editable.
Same pricing as Opus 4.6. If you're already paying for Claude, you just got a significant upgrade for free.
Anthropic quietly retook the lead this week.
Claude Opus 4.7 launched on April 16 with the kind of upgrade that doesn't make for a flashy demo but changes what you can actually build with it. Sharper vision. Stronger coding. Better agentic execution. And a design tool that nobody saw coming from an AI company.
If you use Claude for anything, from writing code to reviewing designs to building products, here's what actually changed.
The benchmarks, honestly
Opus 4.7 leads on most things that matter for people who build software and products. Here's how it stacks up against GPT-5.4 and Gemini 3.1 Pro.
Coding and software engineering:
Benchmark | Claude Opus 4.7 | GPT-5.4 | Gemini 3.1 Pro |
SWE-bench Verified | 87.6% | — | — |
SWE-bench Pro | 64.3% | 57.7% | 54.2% |
CursorBench | 70% | — | — |
Opus 4.7 scores 87.6% on SWE-bench Verified, up from 80.8% on Opus 4.6. That's not a marginal bump. On SWE-bench Pro, it leads GPT-5.4 by nearly seven points.
Knowledge and reasoning:
Benchmark | Claude Opus 4.7 | GPT-5.4 | Gemini 3.1 Pro |
GPQA Diamond | 94.2% | 94.4% | 94.3% |
GDPVal-AA (Elo) | 1,753 | 1,674 | 1,314 |
BrowseComp | 79.3% | 89.3% | 85.9% |
Graduate-level reasoning is a dead heat across all three. Knowledge work (GDPVal-AA) is where Opus 4.7 pulls ahead significantly. Web search (BrowseComp) is the one area where it actually dropped from Opus 4.6, and GPT-5.4 leads there.
Pricing:
Model | Input (per 1M tokens) | Output (per 1M tokens) |
Claude Opus 4.7 | $5 | $25 |
GPT-5.4 | — | — |
Gemini 3.1 Pro | $2 | $12 |
Same pricing as Opus 4.6. No premium for the upgrade.
The honest summary: Opus 4.7 is the strongest model for coding and agentic work. Reasoning is a tie across the board. Web search is its weakest spot. If you're building software or running complex multi-step tasks, it's the best option right now.
Vision that actually works
This is the upgrade that matters more than the benchmarks suggest.
Opus 4.7 accepts images at 3.75 megapixels, up from 1.15 megapixels on Opus 4.6. In practice, that means screenshots, design mockups, dense diagrams, and documents come through at actual fidelity instead of being downscaled to the point where details get lost.
On visual navigation benchmarks, Opus 4.7 scores 79.5% without tools, compared to 57.7% for Opus 4.6. That's a 22-point jump in its ability to understand what it's looking at.
Why this matters beyond benchmarks: you can now share a full-resolution screenshot of your product and get feedback that references specific elements on the page. Not vague suggestions about "improving hierarchy" but actual observations about spacing, contrast, and layout based on what the model can see.
What's new for developers
A few things worth knowing if you're building with Claude.
xhigh effort level. A new setting that sits above the previous effort options. Claude Code defaults to xhigh for all subscriber plans. It gives the model more room to reason through complex problems before responding.
Task budgets (beta). You can set a hard token ceiling on an agentic loop. The model uses a running countdown to prioritize work, which means it won't burn through your budget on a tangent and run out before finishing what matters.
/ultrareview. A new Claude Code command that spawns multiple specialized agents, one for security, one for logic, one for performance, one for style, and synthesizes their findings into a single code review report. Instead of one pass through your codebase, you get four perspectives merged into one.
Adaptive thinking. The model automatically adjusts reasoning depth based on task complexity. Simple questions get fast answers. Hard problems get deeper chains of thought. You don't have to manually toggle between speed and quality.
The design tool nobody expected
Alongside Opus 4.7, Anthropic shipped a design tool that generates websites, landing pages, and presentations from natural language prompts. You describe what you want in plain text and it builds it.
This competes with Gamma, Google Stitch, Canva, and parts of what Figma and Adobe offer. Design-adjacent stocks dipped on the announcement.
The more interesting detail: Anthropic has a partnership with Figma called Code to Canvas, which converts Claude-generated code into fully editable Figma design files. The output isn't a frozen image or a locked template. It's real layers and components that a designer can select, restyle, and build on.
This means the workflow isn't "AI replaces designer." It's "AI generates the starting point, designer refines it in the tool they already use."
What this means for designers specifically
The 3.75-megapixel vision upgrade is the part designers should care about most.
You can now share a screenshot of your homepage with Claude and get specific, layout-aware feedback. Paste a design mockup and ask what's off. Share a competitor's landing page and ask for a breakdown of why it converts better. Show it your brand guidelines and ask whether a generated page matches.
Previous versions were essentially squinting at your designs through a low-resolution window. Opus 4.7 reads the screen closer to how a designer reads the screen.
Combined with the design tool and Figma integration, the workflow looks like this: generate three landing page variations from a prompt, open them in Figma, and spend your time on judgment and refinement instead of building the first draft from scratch. The boring part gets faster. The part that requires taste and business context stays yours.
For founders working with designers: this doesn't change who you need. It changes what they spend their time on. Less blank-canvas construction. More "this is close, here's what needs to change and why."
Should you switch or upgrade?
If you're already on Claude, you got the upgrade automatically. Same price, better model.
If you're deciding between models right now: Opus 4.7 for coding, agentic tasks, and anything involving visual input. GPT-5.4 if web search accuracy is your priority. Gemini 3.1 Pro if cost matters more than peak performance.
For most people building products, shipping code, or working with design, Opus 4.7 is the strongest option available today.
If you're exploring how AI tools like Claude fit into your design and product workflow, that's the kind of conversation I like having. heypash.com
