Claude Opus 4.7 is the newest and most capable publicly available AI model from Anthropic, released on April 16, 2026. It’s a direct upgrade to its predecessor, Opus 4.6, and it brings significant improvements in software engineering, agentic reasoning, visual understanding, and long-running task performance. While Anthropic’s even more powerful Claude Mythos Preview model exists behind closed doors, Opus 4.7 represents the cutting edge of what developers and everyday users can actually get their hands on today.
In this article, we’ll discuss what makes Claude Opus 4.7 a meaningful step forward, how it stacks up against competitors like OpenAI’s GPT-5.4 and Google’s Gemini 3.1 Pro, what new features it introduces, and what its release signals about the broader direction of AI development. Whether you’re a developer evaluating your next model upgrade or simply curious about the state of the art, this breakdown covers everything you need to know.
TL;DR Snapshot
Claude Opus 4.7 is Anthropic’s most capable generally available model as of April 2026. It delivers major gains in coding benchmarks, introduces a new “xhigh” reasoning effort level, triples the supported image resolution, and ships with built-in cybersecurity safeguards tied to Anthropic’s Project Glasswing initiative. Pricing remains unchanged from Opus 4.6 at $5 per million input tokens and $25 per million output tokens.
Key takeaways include…
- Benchmark-leading coding performance: Opus 4.7 scores 64.3% on SWE-bench Pro and 87.6% on SWE-bench Verified, outperforming both GPT-5.4 and Gemini 3.1 Pro on the software engineering tasks that matter most to developers, according to Anthropic’s official announcement.
- Self-verification and stronger autonomy: The model can now check its own outputs before reporting back, a behavior that early-access partners like Vercel and Notion have highlighted as a practical game-changer for long-running agentic workflows.
- New cybersecurity safeguards: Anthropic is using Opus 4.7 as a testbed for safety mechanisms it plans to eventually apply to its more powerful Mythos-class models, including automated detection and blocking of high-risk cybersecurity requests.
Who should read this: Software engineers, AI product builders, enterprise decision-makers, and AI enthusiasts.
A Major Leap in Coding and Software Engineering
The headline story of Opus 4.7 is its performance on coding benchmarks. According to Anthropic’s launch blog post, the model scores 87.6% on SWE-bench Verified (up from 80.8% on Opus 4.6) and 64.3% on SWE-bench Pro (up from 53.4%). That SWE-bench Pro jump of nearly 11 percentage points in a single release is especially notable because it measures harder, multi-language engineering tasks that are more representative of real-world production work.
On CursorBench, which evaluates autonomous coding quality inside the popular Cursor editor, Opus 4.7 scored 70%, up from 58% on its predecessor. As Cursor Co-Founder and CEO Michael Truell noted in Anthropic’s announcement, the model represents “a meaningful jump in capabilities” with “more creative reasoning.” Rakuten, another early-access partner, reported that Opus 4.7 resolved three times more production tasks compared to Opus 4.6.
Users have also reported that Opus 4.7 follows instructions more literally than previous models. Anthropic itself flags this as both a strength and a migration consideration: prompts that relied on the model loosely interpreting vague instructions may now produce different results because Opus 4.7 takes wording more precisely at face value.
New Features: xhigh Effort, Better Vision, and Self-Verification
Beyond raw benchmarks, Opus 4.7 introduces several practical features aimed at giving developers finer control over how the model operates.
The most prominent addition is the new xhigh effort level, which sits between the existing high and max settings. As Axios reported, Anthropic described xhigh as giving users “finer control over the tradeoff between reasoning and latency on hard problems.” Claude Code, Anthropic’s command-line coding agent, now defaults to xhigh for all plans. Anthropic is also beta-testing a feature called “task budgets” that lets developers set limits on how much reasoning the model does during longer tasks.
On the vision front, Opus 4.7 supports images up to 2,576 pixels on the long edge, roughly 3.75 megapixels. That’s more than three times the resolution supported by prior Claude models. According to Anthropic’s announcement, early-access partner Solve Intelligence highlighted how the higher resolution is helping them build better tools for interpreting technical diagrams and chemical structures in life sciences patent work.
Perhaps the most consequential behavioral change is self-verification. Opus 4.7 actively checks its own work before declaring a task complete. It writes tests, runs sanity checks, and inspects its output. According to Notion AI Lead Sarah Sachs, this kind of reliability improvement is what “makes Notion Agent feel like a true teammate.” Early-access partner Intuit described the model as “catching its own logical faults during the planning phase.”
Competitive Landscape: Opus 4.7 vs. GPT-5.4 and Gemini 3.1 Pro
Opus 4.7’s release comes at a moment when the race among frontier AI models is tighter than ever. According to a review from The Next Web, Opus 4.7 leads GPT-5.4 on SWE-bench Pro (64.3% vs. 57.7%) and on CursorBench (70% vs. lower scores from competitors). However, on graduate-level reasoning (GPQA Diamond), all three frontier models have essentially converged around 94%, suggesting that the competitive differentiation has shifted from raw reasoning to applied, multi-step task performance.
There’s one notable weakness. On BrowseComp, a benchmark that evaluates web research and information synthesis, Opus 4.7 scored 79.3%, trailing GPT-5.4 Pro’s 89.3% and Gemini 3.1 Pro’s 85.9%. For teams building agents that rely heavily on real-time web retrieval, this is worth paying attention to.
On pricing, Anthropic kept Opus 4.7 at the same $5/$25 per million token rate as Opus 4.6. However, as a Finout analysis pointed out, the model uses a new tokenizer that can map the same text to 1.0x to 1.35x more tokens. That means your effective cost per request could increase by up to 35% on certain workloads, particularly code, structured data, and non-English text, even though the sticker price hasn’t changed.
Google’s Gemini 3.1 Pro remains cheaper at $2/$12 per million tokens for input and output respectively, which may matter for cost-sensitive production workloads where coding performance isn’t the top priority.
The Mythos Question and Cybersecurity Safeguards
One of the most interesting dynamics around this release is its relationship to Claude Mythos Preview, Anthropic’s most powerful model, which remains restricted to a select group of companies through Project Glasswing. As CNBC reported, Anthropic openly acknowledged that Opus 4.7 doesn’t match Mythos Preview’s capabilities.
This transparency is deliberate. Anthropic is using Opus 4.7 as a proving ground for cybersecurity safeguards that it hopes to eventually apply to Mythos-class models before a broader release. The company stated that it “experimented with efforts to differentially reduce” Opus 4.7’s cyber capabilities during training. The model ships with automated detection systems that block requests indicating prohibited or high-risk cybersecurity uses.
For security professionals who want to use Opus 4.7 for legitimate purposes like penetration testing, vulnerability research, and red-teaming, Anthropic has launched a new Cyber Verification program. According to CNBC, the launch of Project Glasswing has already prompted high-profile conversations between members of the Trump administration, tech CEOs, and bank executives about the security risks posed by powerful AI models.
Availability and Pricing
Opus 4.7 is available immediately across all Claude products, the Claude API (using the model ID claude-opus-4-7), Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry. According to GitHub’s changelog, the model is also rolling out on GitHub Copilot, where it will eventually replace Opus 4.5 and Opus 4.6 in the model picker for Copilot Pro+ users.
Pricing is unchanged: $5 per million input tokens and $25 per million output tokens, with up to 90% savings through prompt caching and 50% through the Batch API. The model supports a 1 million token input context window with up to 128K output tokens.
It’s worth noting that Amazon Web Services emphasized that Bedrock’s deployment of Opus 4.7 provides zero operator access, meaning customer prompts and responses aren’t visible to either Anthropic or AWS operators.
Frequently Asked Questions
Anthropic is an AI company founded in 2021 that builds the Claude family of AI models. The company focuses on developing AI systems that are safe, steerable, and reliable, and it has positioned itself as a more safety-conscious alternative to competitors like OpenAI and Google DeepMind.
Claude Opus 4.7 is Anthropic’s most capable publicly available AI model, released on April 16, 2026. It excels at software engineering, long-running agentic tasks, and professional knowledge work, and it’s available through Anthropic’s own products as well as cloud platforms like Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
Claude Mythos Preview is Anthropic’s most powerful AI model, but it isn’t available to the general public. It’s currently restricted to a select group of technology and cybersecurity companies through Anthropic’s Project Glasswing initiative. Anthropic has stated that it plans to use lessons from deploying safeguards on models like Opus 4.7 to work toward an eventual broader release of Mythos-class models.
Project Glasswing is Anthropic’s cybersecurity initiative, announced in April 2026, that explores both the risks and benefits of advanced AI models in the cybersecurity domain. It involves limited deployment of Claude Mythos Preview to vetted organizations and the development of safeguards designed to prevent AI models from being misused for cyberattacks.
The xhigh effort level is a new reasoning setting introduced with Opus 4.7. It sits between the existing high and max effort levels and gives developers more granular control over how deeply the model reasons through a problem, balancing quality against response latency. Claude Code now defaults to xhigh for all plans.
SWE-bench is a widely used benchmark for evaluating AI models on software engineering tasks. SWE-bench Verified tests the model’s ability to solve real GitHub issues, while SWE-bench Pro is a harder variant that spans multiple programming languages and more complex engineering challenges.
CursorBench is a benchmark that measures how well AI models perform autonomous coding tasks within Cursor, a popular AI-powered code editor. It’s considered a practical indicator of how useful a model is in the real-world development environment where many professional developers actually work.
Amazon Bedrock is AWS’s managed platform for building AI applications and agents at production scale. It allows enterprise customers to access foundation models from multiple providers, including Anthropic’s Claude, through a unified API with enterprise-grade infrastructure and security features.
Other Enterprise AI Articles You May Be Interested In
What Is Composable AI Decisioning? GrowthLoop’s New Platform Explained
Adobe’s Firefly AI Assistant: A New Era of Agentic Creativity
Open Source Quantum AI Is Here: Everything You Need to Know About NVIDIA Ising
Apple’s AI Glasses Are Coming to Take on Meta Ray-Bans: What You Need to Know
Amazon Trainium Chips for Sale? Jassy Hints at Third-Party Sales as Demand Soars
