Introduction
TL;DR Anthropic made a major move on April 16, 2026. The company launched Claude Opus 4.7, its most capable, generally available AI model. This release was not a quiet update. It came with benchmark results, new developer features, and a clear message. Anthropic built this model for the hardest tasks professionals face every day.
Claude Opus 4.7 sits at the top of Anthropic’s publicly accessible model lineup. It outperforms its predecessor across coding, vision, and complex reasoning. For developers, researchers, and enterprise teams, this model marks a real step forward. This blog covers everything worth knowing about the launch, the numbers, the new features, and who should use this model.
Table of Contents
What Is Claude Opus 4.7
The Model at a Glance
Claude Opus 4.7 is Anthropic’s flagship, generally available language model as of April 2026. The model ID on the Anthropic API is claude-opus-4-7. It replaces Claude Opus 4.6 as the default recommendation for high-complexity workloads.
The model supports a 1 million token context window. It handles up to 128,000 output tokens per call. It works across text and vision inputs. High-resolution image support arrives for the first time, with images up to 2,576 pixels on the long edge — roughly 3.75 megapixels.
Claude Opus 4.7 is available on Claude.ai through Pro, Max, Team, and Enterprise plans. API access covers the Anthropic platform, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, and GitHub Copilot for Pro+ and Business tiers.
What Makes This a Meaningful Upgrade
Claude Opus 4.7 is not a cosmetic version bump. Anthropic made targeted improvements in the areas that matter most for real production use. Coding performance jumped significantly. Vision quality improved by a large margin. Agentic reliability improved across multi-step and long-horizon tasks.
The model also introduces new developer controls, including a new effort level called xhigh. This gives teams finer control over how much reasoning depth the model applies. More on that in the features section.
Claude Opus 4.7 Benchmark Performance
Coding Benchmarks
Coding is where Claude Opus 4.7 makes its strongest statement. On SWE-bench Verified, the curated benchmark of 500 human-validated GitHub issues, the model scores 87.6%. Opus 4.6 scored 80.8% on the same test. That is a 6.8 percentage point gain in a single version.
SWE-bench Pro is an even harder subset. Claude Opus 4.7 scores 64.3% there, up from 53.4% on Opus 4.6. That is nearly an 11-point jump. CursorBench, which tests AI coding performance inside the Cursor IDE, moved from 58% to 70%. These are not incremental changes.
Early enterprise testing from the data analytics company Hex showed a 13% lift in task resolution on their internal 93-task coding benchmark. The evaluation also showed Opus 4.7 solved four problems that neither Opus 4.6 nor Sonnet 4.6 could crack.
Vision Benchmarks
Claude Opus 4.7 marks a turning point for vision. The maximum image resolution jumped from 1,568 pixels to 2,576 pixels on the long edge. In real terms, that means 3.75 megapixels versus 1.15 megapixels before. This matters for dense diagrams, design mockups, and high-detail screenshots.
On visual navigation tasks without tools, Opus 4.7 scores 79.5%. Opus 4.6 scored 57.7% on the same tests. That is a 22-point improvement. Visual acuity scores reached 98.5%, up from 54.5% on Opus 4.6. Vision has become a genuine first-class capability, not a secondary feature.
Knowledge Work and Reasoning
On GPQA Diamond, which tests scientific reasoning across graduate-level questions, Claude Opus 4.7 scores 94.2%. On GDPVal-AA, an Elo-based knowledge work benchmark, the model scores 1,753. GPT-5.4 scores 1,674 on the same benchmark. Gemini 3.1 Pro scores 1,314. That last gap is decisive.
Claude Opus 4.7 also scores 64.4% on Finance Agent, a benchmark focused on structured financial reasoning tasks. Enterprise teams working on financial modeling, legal analysis, or document-heavy workflows will see that number.
Cybersecurity Benchmarks
On CyberGym, which tests vulnerability reproduction, Claude Opus 4.7 scores 73.1%. The restricted Claude Mythos Preview scores 83.1% on the same test. GPT-5.4 scores 66.3%. Anthropic intentionally reduced Claude Opus 4.7’s raw cyber capabilities during training. The company simultaneously launched a Cyber Verification Program for legitimate security professionals to access the full range of those capabilities for defensive work.
Where the Model Falls Short
Honest reporting matters here. Claude Opus 4.7 regresses on Terminal-Bench 2.0, scoring 69.4% versus GPT-5.4’s 75.1%. It also softens slightly on BrowseComp compared to Opus 4.6. Teams running terminal-heavy workflows or extensive web browsing agents should note this before switching.
Key New Features in Claude Opus 4.7
The xhigh Effort Level
Claude Opus 4.7 introduces a new effort level called xhigh. This slots between the existing high and max levels. The full effort ladder now reads: low, medium, high, xhigh, max.
This gives developers finer-grained control over reasoning depth. Before this release, teams either used high for most tasks or paid the full latency cost of max for difficult ones. The xhigh level fills that gap. It extracts more thinking from the model without incurring the maximum possible latency.
For practical use, xhigh works well on tasks that need careful multi-step reasoning but not the full deliberation of max mode. Legal document review, complex data transformations, and structured financial analysis are good candidates.
Adaptive Thinking
Claude Opus 4.7 supports adaptive thinking. This feature lets the model dynamically allocate thinking token budgets based on the complexity of each individual request. Simple queries get less thinking budget. Hard problems get more. The model decides this automatically.
AWS Bedrock users get this feature natively through the Anthropic Messages API. The result is more intelligent resource use across mixed workloads where task complexity varies request by request.
The /ultrareview Command in Claude Code
Claude Code users get a new /ultrareview command with Claude Opus 4.7. Standard code review checks for syntax errors and obvious bugs. The /ultrareview command does something different. It simulates a senior human reviewer. It flags subtle design flaws, architectural issues, and logic gaps that surface during planning phases rather than execution.
GitHub’s early testing described the command as a meaningful improvement for catching issues before they reach production. For teams using Claude Code in CI/CD pipelines, this command adds a new quality gate.
Task Budgets (Beta)
Task budgets arrive in beta with Claude Opus 4.7. This feature lets teams set computational and resource constraints on individual tasks or workflows. It gives production engineers better cost predictability. When a task approaches its budget, the model can either complete within constraints or surface a clear signal that the task needs more resources.
This is especially useful in agentic workflows where long-running tasks can quietly consume large amounts of compute before an engineer notices.
Auto Mode for Max Plan Users
Auto mode — where Claude can make autonomous decisions without constant permission prompts — now extends to Max plan users. Previously this was more restricted. For Max plan subscribers, Claude Opus 4.7 can operate more independently across multi-step agentic tasks. This streamlines workflows where constant human approval creates unnecessary friction.
High-Resolution Vision
As covered in the benchmarks section, Claude Opus 4.7 supports images up to 2,576 pixels on the long edge. This is the first Claude model with what Anthropic describes as high-resolution image support. Dense technical diagrams, financial charts, design mockups, and screenshots of complex interfaces become far more legible.
Claude Opus 4.7 Pricing
Pricing Details
Claude Opus 4.7 pricing stays at $5 per million input tokens and $25 per million output tokens. This matches Opus 4.6 exactly. For teams already budgeting for Opus 4.6, the per-token price does not change.
The important caveat is the updated tokenizer. Claude Opus 4.7 uses a new tokenizer that processes the same text into roughly 1.0 to 1.35 times more tokens than Opus 4.6. This means the same input can cost between zero percent and 35 percent more, depending on content type.
Teams migrating from Opus 4.6 at scale should benchmark their specific inputs before committing. The headline price is unchanged. The effective cost per request may increase depending on content.
Caching and Batch Processing
Anthropic’s standard caching and batch processing discounts apply to Claude Opus 4.7. For teams running high-volume, repetitive prompts, prompt caching can substantially reduce costs. Batch processing, where requests are queued for non-real-time processing, also offers cost reductions. Both features carry over from Opus 4.6 with no configuration changes required.
Claude Opus 4.7 vs Competitors
Claude Opus 4.7 vs GPT-5.4
Claude Opus 4.7 and GPT-5.4 are the two strongest generally available models in April 2026. On coding benchmarks, Claude Opus 4.7 leads. On SWE-bench Verified, Claude Opus 4.7 scores 87.6% versus GPT-5.4’s lower score. On CyberGym, Claude Opus 4.7 scores 73.1% against GPT-5.4’s 66.3%.
GPT-5.4 leads on Terminal-Bench 2.0 with 75.1% against Opus 4.7’s 69.4%. For knowledge work measured on GDPVal-AA, Claude Opus 4.7 scores 1,753 versus GPT-5.4’s 1,674. The coding and knowledge work gap favors Claude Opus 4.7.
Claude Opus 4.7 vs Claude Mythos Preview
Claude Mythos Preview is a more powerful model than Claude Opus 4.7. Anthropic has acknowledged this clearly. Mythos discovered thousands of high-severity zero-day vulnerabilities. On every benchmark, Mythos Preview leads Opus 4.7. However, Mythos Preview is not generally available. Anthropic launched it to a select group of companies through Project Glasswing, a cybersecurity research initiative. Claude Opus 4.7 is the strongest model anyone can actually access and use today.
Claude Opus 4.7 vs Claude Opus 4.6
The comparison is straightforward. Claude Opus 4.7 outperforms Opus 4.6 on coding, vision, knowledge work, agentic reliability, and instruction following. SWE-bench Verified improved by 6.8 points. Vision acuity improved by 44 points. CursorBench improved by 12 points. The only regression areas are Terminal-Bench 2.0 and BrowseComp. For most teams, Claude Opus 4.7 is a clear upgrade. Anthropic has announced the deprecation of Claude Opus 4.6, with retirement scheduled for June 15, 2026.
Who Should Use Claude Opus 4.7
Software Engineering Teams
Claude Opus 4.7 is the best choice for software engineering teams running agentic coding workflows. The 87.6% score on SWE-bench Verified sets the state of the art for publicly available models. The /ultrareview command in Claude Code adds senior-level review to automated pipelines. Teams doing complex refactoring, debugging, and long-horizon task execution will see meaningful gains.
Enterprise Knowledge Workers
Legal teams, financial analysts, and research professionals handle dense, complex documents. Claude Opus 4.7 leads on GDPVal-AA knowledge work benchmarks with a score of 1,753. It scores 64.4% on Finance Agent. The 1 million token context window means it can hold an entire large report, contract, or codebase in memory during a single session.
Vision-Dependent Workflows
Any team working with technical diagrams, design mockups, high-density charts, or screenshots will benefit from the upgraded vision capabilities. The jump from 1.15 megapixels to 3.75 megapixels is not marginal. It fundamentally changes what Claude Opus 4.7 can extract from visual inputs. Interface designers, data analysts, and documentation reviewers are natural users of this capability.
Agentic AI Application Developers
Developers building autonomous AI agents need reliable, multi-step execution. Claude Opus 4.7 delivers stronger performance on long-horizon agentic tasks than any previous Anthropic model. The adaptive thinking feature, task budgets, and auto mode for Max plan users all support complex autonomous workflows. The self-verification behavior, where the model catches its own logical faults during planning, reduces error propagation across extended task sequences.
Security Professionals
Legitimate security professionals — vulnerability researchers, penetration testers, and red-teamers — should look at the Cyber Verification Program Anthropic launched alongside Claude Opus 4.7. This program provides verified access to stronger cybersecurity capabilities for defensive work. The 73.1% CyberGym score makes Claude Opus 4.7 competitive with GPT-5.4 for security research tasks.
How to Access Claude Opus 4.7
Claude.ai
Claude Opus 4.7 is available on Claude.ai for Pro, Max, Team, and Enterprise plan subscribers. Pro users can access the model directly. Max plan users get auto mode enabled. Enterprise plan subscribers get priority access and usage controls.
Anthropic API
The model identifier on the Anthropic API is claude-opus-4-7. Teams migrating from Opus 4.6 should update their model string. The API otherwise maintains the same interface. Caching, tool use, multi-turn conversations, and streaming all work as before. Teams should also account for the tokenizer change when estimating costs for existing prompts.
Amazon Bedrock
Claude Opus 4.7 is available on Amazon Bedrock in 27 AWS regions. Both the Bedrock-native Converse API and the Anthropic Messages API are supported. Bedrock’s new inference engine provides dynamic capacity allocation and zero operator access, meaning customer prompts never reach Anthropic or AWS operators.
Google Cloud Vertex AI
Google Cloud customers can access Claude Opus 4.7 through Vertex AI. The same API interface and model capabilities are available. Teams already using Vertex AI for other Claude models can switch the model string to claude-opus-4-7.
Microsoft Foundry and GitHub Copilot
Microsoft Foundry supports Claude Opus 4.7 for enterprise API use. GitHub Copilot users on Pro+, Business, and Enterprise plans will see Claude Opus 4.7 replace Opus 4.5 and Opus 4.6 in the model picker over the coming weeks.
Frequently Asked Questions About Claude Opus 4.7
What is Claude Opus 4.7?
Claude Opus 4.7 is Anthropic’s most capable, generally available AI model as of April 16, 2026. It improves on Claude Opus 4.6 across coding, vision, knowledge work, and agentic task execution. It supports a 1 million token context window and high-resolution image inputs.
When did Anthropic release Claude Opus 4.7?
Anthropic released Claude Opus 4.7 on April 16, 2026. It became generally available across Claude.ai, the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry on the same day.
How much does Claude Opus 4.7 cost?
Claude Opus 4.7 pricing is $5 per million input tokens and $25 per million output tokens. This matches Opus 4.6 pricing. Teams should note the updated tokenizer can produce 1.0 to 1.35 times more tokens from the same input, which may increase effective per-request costs.
What is the model ID for Claude Opus 4.7?
The API model identifier is claude-opus-4-7. Teams migrating from Opus 4.6 should update their code to reference this model string.
How does Claude Opus 4.7 compare to Claude Opus 4.6?
Claude Opus 4.7 leads Opus 4.6 across nearly every benchmark. SWE-bench Verified improved from 80.8% to 87.6%. Vision acuity improved from 54.5% to 98.5%. CursorBench improved from 58% to 70%. Terminal-Bench 2.0 is the only notable regression, where Opus 4.7 scores 69.4% versus Opus 4.6’s stronger showing.
Is Claude Opus 4.7 available on Amazon Bedrock?
Yes. Claude Opus 4.7 is available on Amazon Bedrock in 27 AWS regions. It works through both the Bedrock-native Converse API and the Anthropic Messages API.
What is the xhigh effort level in Claude Opus 4.7?
The xhigh effort level is a new reasoning depth control introduced with Claude Opus 4.7. It sits between the existing high and max levels on the effort ladder. It gives teams more reasoning depth than high without the full latency cost of max. It suits tasks requiring careful multi-step reasoning but not maximum deliberation.
What is the /ultrareview command?
The /ultrareview command is available in Claude Code for Claude Opus 4.7 users. It simulates a senior human code reviewer, identifying subtle design flaws, architectural problems, and logic gaps that standard review tools often miss. It adds a deeper quality layer to automated development pipelines.
Is Claude Mythos Preview better than Claude Opus 4.7?
Yes. Claude Mythos Preview outperforms Claude Opus 4.7 on every known benchmark. However, Mythos Preview is not generally available. Claude Opus 4.7 is the strongest model anyone outside Anthropic can access today.
Will Claude Opus 4.6 be deprecated?
Yes. Anthropic has announced the deprecation of Claude Opus 4.6 with retirement scheduled for June 15, 2026. Anthropic recommends migrating to Claude Opus 4.7 before that date.
Read More:-LangGraph Tutorial for Beginners
Conclusion
Claude Opus 4.7 is Anthropic’s clearest statement yet about what a production-ready, frontier AI model looks like. The improvements are real, well-documented, and directly relevant to the workflows enterprise teams run every day.
Software engineers get the strongest publicly available coding model. Vision-heavy workflows get a capability that previously required workarounds. Knowledge workers get a model that outpaces competitors on demanding reasoning benchmarks. Developers building agents get better agentic reliability, smarter effort controls, and stronger self-verification behavior.
The pricing story is mostly good news. The per-token rate stays the same. The tokenizer change requires monitoring for cost-sensitive teams. The performance gain at unchanged pricing makes Claude Opus 4.7 a compelling upgrade.
This release matters beyond the numbers. Anthropic built this model for the hardest tasks professionals encounter. The results across coding, vision, and knowledge work benchmarks show that ambition translated into execution. Claude Opus 4.7 sets a new benchmark for what generally available AI can do in April 2026.
