Anthropic prepares Opus 4.7: the AI design tool that changes everything

On April 15, 2026, The Information quietly dropped a bombshell: Anthropic is gearing up for the imminent launch of Claude Opus 4.7, already spotted in internal API references. This isn’t just an update. Opus 4.7 comes with an AI design tool capable of generating websites, landing pages, and presentations through natural language. On the same day, Adobe, Figma, and Wix shares fell by over 2%. The message is clear.

What Opus 4.7 really is and why it matters

Claude Opus 4.7 follows Opus 4.6, launched on February 5, 2026, and currently the most advanced model in the Claude range available to the public. Sixty-eight days separate the two releases, confirming the accelerated monthly pace Anthropic has maintained since late 2025.

The model addresses several weaknesses felt in Opus 4.6: less precise answers, inconsistent reasoning, and query limits reached too quickly. To fix this, Opus 4.7 injects a massive volume of tokens server-side, deepening analysis without burdening user input. Leaks already show concrete results: generating complete UI interfaces with much higher consistency than previous versions, like accurately reproducing a GPT-6 main menu from a simple screenshot.

In terms of architecture, Opus 4.7 introduces a restructuring of the Model Context Protocol (MCP) to reduce latency in agent loops. The Extended Thinking mode with stateful memory allows building a complete mental map of a codebase. It’s no longer just an enhanced chatbot: it’s a tool designed for autonomous production.

To understand this trajectory, revisit our analysis of Claude Opus 4.6 and its uses in finance and legal, which already laid the groundwork for this rise in power.

The AI design tool: the real bombshell of the launch

Opus 4.7 doesn’t launch alone. Anthropic pairs it with a web and presentation design tool entirely driven by natural language. The principle is simple: you describe what you want, the tool generates the result. No coding, no need to master Figma or Adobe.

The target audiences are clear:

• Presentations: direct competitor to Gamma and Google Slides
• Websites and landing pages: competitor to Wix, Squarespace, and Webflow
• UI interfaces: competitor to Figma and Adobe XD

A concrete example from the leaks: a non-developer can generate a complete e-commerce site by simply describing their needs. A freelance designer can create a portfolio in minutes through prompts. This positioning targets both non-technical users and pros looking to speed up their workflow.

Warning: the market impact is already visible. The April 15, 2026 leak caused a drop of over 2% in Adobe, Figma, and Wix shares in a single session. If Anthropic delivers on its promises at launch, the pressure on these players will intensify significantly.

The tool fits into an integrated ecosystem that Anthropic is methodically building: Claude Opus 4.7 as the engine, a revamped Claude Code Desktop as the development environment, a full-stack AI Studio for building/testing/deploying without leaving the platform. All this against competitors like Lovable or Replit.

The Claude model hierarchy in April 2026

A commonly misunderstood point: Opus 4.7 is not Anthropic’s most powerful model. That title goes to Claude Mythos, in preview testing since April 2026 with selected partners for security vulnerability research. It has a dedicated system card and remains inaccessible to the general public.

The current hierarchy is structured as follows:

On the WebDev Arena benchmarks, Claude’s dominance is overwhelming: five of the top six spots are held by Claude models. Opus 4.6 Thinking leads with an Elo score of 1548, ahead of Opus 4.6 standard at 1542. GPT-5.4 High ranks 7th at 1457. With Opus 4.7, this advantage is expected to widen in reasoning and consistency.

This dynamic explains why Anthropic captivates investors so much. To delve deeper into the platform’s agent capabilities, read our report on Anthropic Computer Use.

Record valuation and new pricing model: what changes for you

Anthropic’s figures at the end of 2025-beginning of 2026 are staggering. The annualized revenue reaches $30 billion, up from $9 billion just months earlier. More than 1,000 enterprise clients each spend over a million dollars annually, a figure that doubled in less than two months. Platforms like AWS Bedrock and Google Vertex AI integrate Claude on a large scale.

In this context, investors value Anthropic at up to $800 billion, more than double the valuation achieved during the February 2026 funding round. This estimate is directly linked to the announcement of Opus 4.7 and the design tool.

Tip: If you use Claude intensively via API, prepare your budget. Anthropic is shifting to a usage-based billing model for enterprises, replacing the fixed rate. A heavy-user client could see their bill rise from $1 million to $3 million annually. Assess your token volume before the Opus 4.7 launch.

This economic model change is explained by the increased computing costs. Opus 4.7 consumes more server resources than Opus 4.6 due to the massive token injection. Access to the model will remain available via Claude API, AWS Bedrock, and Google Vertex AI, with a likely model identifier like claude-opus-4-7, similar to the current convention.

Comparison with OpenAI is inevitable. In terms of reasoning and consistency, Opus 4.7 outpaces GPT according to leaks. In pure speed, OpenAI retains advantages in certain use cases. But the bundle Anthropic is building around Opus 4.7 (model + Code Desktop + design tool + AI Studio) changes the competitive landscape. It’s a platform, not an isolated model. To grasp the full scope of this battle, our comparison OpenAI Operator vs Anthropic Computer Use remains a must-read.

What Opus 4.6’s performance reveals about Opus 4.7

To anticipate what Opus 4.7 will bring, Opus 4.6’s figures are instructive. On high-difficulty benign query evaluations, Opus 4.6’s over-refusal rate drops to 0.04%, compared to 8.50% for Sonnet 4.5 and 6.01% for Haiku 4.5. In short: the model answers your legitimate questions without excessive suspicion, even on sensitive topics.

Opus 4.6 supports text and image input, with text output up to 128,000 tokens, a reliable knowledge cutoff in May 2025, and training data up to August 2025. Its hybrid reasoning architecture, combined with RLHF and AI feedback, gives it an edge in extended autonomous agent tasks, especially in coding.

Opus 4.7 is expected to push these parameters further, with:

1. Better prompt response accuracy for complex queries
2. Increased generational consistency over long sessions
3. Superior realistic details in visual and textual outputs
4. Improved coordination between AI agents on multi-step tasks

Security remains a central concern at Anthropic. Constitution AI guides responses towards useful, honest, and harmless behaviors. But the line between security and alignment is sometimes finer than it seems. On this topic, our article on AI’s survival instinct and alignment limits provides essential insights.

Conclusion

Opus 4.7 isn’t just a version update. It’s the pivot of a product strategy transforming Anthropic into a complete creation platform. A more precise model, a natural language design tool, a revamped Claude Code Desktop, a full-stack AI Studio: Anthropic is building a closed and coherent environment, designed to attract users from Adobe, Figma, Wix, and other Gamma tools.

The launch is imminent according to mid-April 2026 indications, with full deployment expected by the end of Q2. For businesses, the shift to usage-based billing deserves immediate attention. For creatives and developers, Opus 4.7 opens up concrete opportunities for time and quality gains. Keep a close eye on official announcements on the Anthropic Transparency page to not miss the launch.

FAQ

What is the difference between Claude Opus 4.7 and Claude Opus 4.6?

Opus 4.7 addresses weaknesses felt in Opus 4.6: less precise answers, inconsistent reasoning, and query limits reached too quickly. It injects a high volume of tokens server-side to deepen analysis without burdening input. It also comes with an AI design tool and a revamped Claude Code Desktop, absent from version 4.6.

Is Opus 4.7 the most powerful model from Anthropic?

No. Claude Mythos holds that position. It has been in preview testing since April 2026, reserved for selected partners for security vulnerability research. Opus 4.7 is the flagship model widely accessible, aimed at businesses and developers via API, AWS Bedrock, and Google Vertex AI.

How to access Claude Opus 4.7 once launched?

Access will be through Claude API with a dedicated model identifier, as well as via AWS Bedrock and Google Vertex AI for enterprises. The Claude Pro subscription will likely provide access to the model on claude.ai, as with previous versions. Billing is shifting to a usage-based model for enterprise clients.

Can Anthropic’s AI design tool replace Figma or Adobe?

It targets the same uses: creating websites, landing pages, presentations, UI interfaces, via natural language instructions. For a non-developer or a creative freelancer, it offers a quick and accessible alternative. For complex professional workflows, Figma and Adobe retain advantages in precision and advanced collaboration. The 2% drop in these publishers’ shares on April 15, 2026, shows that the market perceives the threat as real.

What impact will the shift to usage-based billing have on costs?

For occasional users, the impact will be limited. For heavy enterprise users, costs could triple: a client going from $1 million to $3 million annually according to available estimates. This change is directly linked to the increased computing costs caused by Opus 4.7 and its massive server tokens. It’s advised to assess your usage volume before the official launch.