TL;DR Opus 4.7 is a targeted upgrade over 4.6, improving coding, vision, and agentic reliability without changing pricing. It follows instructions more strictly, handles multi-step tasks better, and supports higher-resolution inputs—though token usage may increase slightly due to a new tokenizer.

Anthropic released Claude Opus 4.7 on April 16, 2026, and it arrives without the usual fanfare that tends to follow a new model drop. There is no claim that it will replace your engineering team. No suggestion it is "the smartest AI ever built." Instead, Opus 4.7 reads like a deliberate refinement of what came before, a model that is better at the work teams were already trying to hand off to its predecessor.

If you are building with Claude, here is what Opus 4.7 actually brings to the table and what you should know before switching.

Claude Opus 4.7 vs 4.6: Technical Comparison

Before we get into the details, here is a side-by-side look at how Claude Opus 4.7 compares to its predecessor.

Feature	Opus 4.6	Opus 4.7
SWE-bench Verified	80.8%	87.6%
Max Image Resolution	~1.1 MP	3.75 MP (2,576px long edge)
Best For	Stable production	Coding agents and complex vision
Instruction Following	Fluid, interpretive	Literal, strict
Effort Controls	High / Max	New "xhigh" effort level
Pricing (Input / Output)	$5 / $25 per 1M tokens	$5 / $25 per 1M tokens
Tokenizer	Standard	Updated (1.0 to 1.35× token increase)
Computer Use	Supported	Supported, with 3× higher resolution

The pricing has not changed. But the model's behaviour has, in ways that matter for production workloads. Let us walk through each one.

What Is New in Claude Opus 4.7

Claude Opus 4.7 improves on its previous release, Opus 4.6, in a few specific, well-defined areas. It is not a generational leap. It is a targeted upgrade that addresses the friction points developers and teams have been hitting in production.

Coding and agentic work

The headline improvement is in software engineering, particularly on difficult, long-running tasks. Where Opus 4.6 sometimes needed close supervision during complex coding sessions, Claude Opus 4.7 handles multi-step work with more consistency. It verifies its own outputs before reporting back, which means fewer "looks right, is not right" moments.

Early testers reported meaningful gains. Cursor saw its benchmark scores jump from 58% to over 70%. Rakuten's engineering team reported that Opus 4.7 resolves three times more production tasks than its predecessor. Notion's AI team measured a 14% improvement over Opus 4.6 with fewer tokens consumed and a third of the tool errors.

Instruction following that actually follows

Claude Opus 4.7 takes instructions more literally than previous models. That sounds like a small thing. It is not. Earlier Claude models would sometimes interpret prompts loosely or skip parts entirely. Opus 4.7 treats your prompt like it means what it says.

Anthropic notes that prompts written for earlier models may now produce unexpected results because the model no longer glosses over ambiguity. If you are migrating, re-tuning prompts is worth the time.

Cladue Opus 4.7 and The Competitive Landscape

Benchmark	Opus 4.7	Opus 4.6	GPT-5.4	Gemini 3.1 Pro	Mythos Preview
Agentic coding (SWE-bench Pro)	64.3%	53.4%	57.7%	54.2%	77.8%
Agentic coding (SWE-bench Verified)	87.6%	80.8%	—	80.6%	93.9%
Agentic terminal coding (Terminal-Bench 2.0)	69.4%	65.4%	75.1%	68.5%	82.0%
Multidisciplinary reasoning (Humanity’s Last Exam, no tools)	46.9%	40.0%	42.7%	44.4%	56.8%
Multidisciplinary reasoning (Humanity’s Last Exam, with tools)	54.7%	53.3%	58.7%	51.4%	64.7%
Agentic search (BrowseComp)	79.3%	83.7%	89.3%	85.9%	86.9%
Scaled tool use (MCP-Atlas)	77.3%	75.8%	68.1%	73.9%	—
Agentic computer use (OSWorld-Verified)	78.0%	72.7%	75.0%	—	79.6%
Agentic financial analysis (Finance Agent v1.1)	64.4%	60.1%	61.5%	59.7%	—
Cybersecurity vulnerability reproduction (CyberGym)	73.1%	73.8%	66.3%	—	83.1%
Graduate-level reasoning (GPQA Diamond)	94.2%	91.3%	94.4%	94.3%	94.6%
Visual reasoning (CharXiv, no tools)	82.1%	69.1%	—	—	86.1%
Visual reasoning (CharXiv, with tools)	91.0%	84.7%	—	—	93.2%
Multilingual Q&A (MMLU)	91.5%	91.1%	—	92.6%	—

Why CLaude Opus 4.7 Is the New Standard for Visual Reasoning

The model accepts images up to 2,576 pixels on the long edge, roughly 3.75 megapixels. That is more than three times the resolution of prior Claude models, and it turns out to be a bigger deal than it sounds on paper.

This matters for computer use agents reading dense screenshots, data extractions from complex diagrams, and any workflow that depends on fine visual detail.

XBOW, an autonomous penetration testing company, reported that its visual acuity benchmark score went from 54.5% with Opus 4.6 to 98.5% with Opus 4.7. That is not an incremental gain. That is a capability that was previously unreliable, becoming nearly flawless.

Solve Intelligence, which builds tools for life sciences patent workflows, highlighted major improvements in reading chemical structures and interpreting complex technical diagrams. For teams building multimodal applications, Claude Opus 4.7 turns vision from a "nice to have" into something you can actually depend on.

Anthropic also describes the model as "more tasteful and creative" for professional outputs like interfaces, slides, and documents. Lovable's CEO noted that Claude Opus 4.7 is the best model they have tested for building dashboards and data-heavy interfaces, with design choices good enough to ship.

Token Inflation: What the Updated Tokenizer Means for Your Costs

Pricing stays the same as Opus 4.6: $5 per million input tokens and $25 per million output tokens. The model is available through the Claude API (model string: claude-opus-4-7), Amazon Bedrock, Google Cloud's Vertex AI, and Microsoft Foundry.

But here is the nuance worth paying attention to. Opus 4.7 uses an updated tokenizer, and the same input can map to roughly 1.0 to 1.35 times more tokens depending on content type. The model also tends to think more at higher effort levels, which produces more output tokens. So while the per token price has not changed, your total spend on the same workload could edge upward.

Anthropic says the net effect is favourable in their own testing because the model solves problems in fewer attempts. But you will want to measure token usage against your own workloads before assuming costs stay flat.

There are a few levers you can pull to manage this: adjusting the effort parameter, setting task budgets (now in public beta), or prompting the model to be more concise. Anthropic has published a migration guide with further advice.

Breaking change: Opus 4.7 no longer supports temperature, top_p, or top_k. Setting any of these to a non-default value will return a 400 error. Developers should remove these parameters from API calls entirely and use the new effort levels to control output behaviour instead.

You get better results for less work, even if individual requests use slightly more tokens.

What Else Launched Alongside Claude Opus 4.7

Anthropic bundled a few related updates with the release:

New "xhigh" effort level: Sits between "high" and "max," giving developers finer control over the reasoning versus latency tradeoff. Claude Code's default is now set to xhigh.
Task budgets (public beta): A way to guide Claude's token spend so it can prioritise work across longer runs on the API.
/ultrareview in Claude Code: A dedicated review session that reads through code changes and flags bugs and design issues. Pro and Max users get three free ultrareviews to try it.
Auto mode for Max users: Claude makes permission decisions on your behalf during longer tasks, reducing interruptions.

Claude Opus 4.7 vs. Mythos: What Is the Difference?

If you have been following Anthropic's releases, you may be wondering where Opus 4.7 sits relative to Claude Mythos Preview, the company's most capable model. The short answer: they serve different purposes, and only one of them is something you can actually use in production today.

Opus 4.7 is the workhorse. It is generally available, priced the same as Opus 4.6, and designed for teams that need reliable, day-to-day performance on coding, document analysis, and agent workflows. It is the model you build products on.

Mythos Preview is the frontier. It remains in limited release while Anthropic tests new safety measures around its more advanced cyber capabilities. Anthropic has stated that Opus 4.7 is the first model to ship with automated cybersecurity safeguards, and what they learn from deploying those safeguards will inform a broader release of Mythos class models down the road.

On alignment, Anthropic's assessment describes Opus 4.7 as "largely well aligned and trustworthy, though not fully ideal." It is modestly better than Opus 4.6 on honesty and prompt injection resistance, and modestly weaker in a few edge cases around harm reduction advice. Mythos Preview still scores highest on alignment evaluations overall.

For most teams, the question is not "Opus 4.7 or Mythos?" It is "Is Opus 4.7 a meaningful upgrade from 4.6?" And the answer, based on the benchmarks and early tester reports, is yes.

What Opus 4.7 Means for Teams Building with Claude

Opus 4.7 is not trying to be everything to everyone. It is a model that does the hard, unglamorous work: long coding sessions, complex document analysis, multi-step agent workflows, all with more reliability than its predecessor.

If you have been using Opus 4.6 and occasionally babysitting it through tough tasks, Opus 4.7 is designed to need less of that. The upgrade is not dramatic. It is just the kind that makes you stop checking over the model's shoulder.

Building the Future with Thesys

While models like Opus 4.7 push the boundaries of raw intelligence, Thesys provides the application layer. Our platform is designed to turn Claude’s reasoning into interactive Generative UI, transforming static AI responses into professional, production-ready interfaces.

Explore Generative UI with Thesys

As we work on bringing full Opus 4.7 support to our workspace, you can explore the Thesys Agent Builder today to see how we’re redefining the way teams build with AI.

Try Thesys Agent Builder

FAQs

What is Opus 4.7?

Opus 4.7 is Anthropic's latest model in the Claude Opus family, released on April 16, 2026. It improves on Opus 4.6 in coding, vision, instruction following, and agentic tasks, while keeping the same per-token pricing.

How much does Opus 4.7 cost?

Opus 4.7 is priced at $5 per million input tokens and $25 per million output tokens, identical to Opus 4.6. However, the updated tokenizer may result in slightly higher token counts for the same input, so total costs could increase modestly depending on your workload.

Is Opus 4.7 better than Opus 4.6?

Yes, across most benchmarks. It shows particular gains in software engineering, vision, and long running agentic tasks. Anthropic notes it is not as broadly capable as Claude Mythos Preview, but it outperforms Opus 4.6 in the majority of tested domains.

Does Opus 4.7 support Computer Use?

Yes. Opus 4.7 supports computer use with three times the image resolution of previous Claude models. This makes it significantly better at reading dense screenshots, navigating complex UIs, and performing tasks that depend on fine visual detail.

Should I update my prompts when switching to Opus 4.7?

Yes. Opus 4.7 follows instructions more literally than earlier models. Prompts that relied on the model filling in gaps or interpreting loosely may produce different results. Anthropic recommends re-tuning prompts and testing against your specific use cases.

Where is Opus 4.7 available?

It is available through the Claude API, Amazon Bedrock, Google Cloud's Vertex AI, Microsoft Foundry, and all Claude consumer products. The API model string is claude-opus-4-7.

What is the difference between Opus 4.7 and Claude Mythos?

Opus 4.7 is generally available and designed for production workloads. Mythos Preview is Anthropic's most capable model but remains in limited release while the company tests new safety measures. Most teams should use Opus 4.7 for day-to-day work and watch for a broader Mythos release later.

Related articles

May 2026 Security Incident

Why We're Open Sourcing OpenUI

AI Copilots: A Complete Guide to Intelligent Assistance

What Are AI Agents: Types, Capabilities, and Real-World Use

ChatKit: What It Is, What It Isn’t, and Where C1 Fits In

PowerPoint Automation Reimagined with C1 Artifacts API

The Future of AI Interfaces is Generative

MCP UI: Making AI Apps Interactive

What Is AG-UI and Why It Matters for AI-Native User Interfaces

How to design AI-Native Conversational Interfaces : From Templates to Generative UI

GPT 5 vs. GPT 4.1

How to Build Generative UI Applications

Implementing Generative Analytics with Thesys and MCP

Evolution of Analytics: From Static Dashboards to Generative UI

Why Generating Code for Generative UI is a bad idea

Building the First Generative UI API: Technical Architecture and Design Decisions Behind C1

How we evaluate LLMs for Generative UI

Generative UI vs Prompt to UI vs Prompt to Design

Generative UI: What It Is and Why It Matters