Products

Anthropic Releases Claude Opus 4.7 with Major Gains in Software Engineering, Vision, and Instruction Following

On April 16, 2026, Anthropic announced the general availability of Claude Opus 4.7, a notable improvement over Opus 4.6 in advanced software engineering—particularly on the hardest tasks—along with substantially better vision capabilities for high-resolution images and more precise instruction following, while remaining less broadly capable than the limited Claude Mythos Preview.

Tech Insights Reporter Apr 16, 2026 8 min read San Francisco

title: "Anthropic Releases Claude Opus 4.7 with Major Gains in Software Engineering, Vision, and Instruction Following" summary: "On April 16, 2026, Anthropic announced the general availability of Claude Opus 4.7, a notable improvement over Opus 4.6 in advanced software engineering—particularly on the hardest tasks—along with substantially better vision capabilities for high-resolution images and more precise instruction following, while remaining less broadly capable than the limited Claude Mythos Preview." category: "Products" author: "Tech Insights Reporter" date: "2026-04-16" readTime: "8 min read" location: "San Francisco" hue: 220 regions: - "us" platforms: - "anthropic" tag: "Model Release" featured: false

TLDR

Anthropic released Claude Opus 4.7 on April 16, 2026, as its latest generally available flagship model. The release emphasizes gains in complex, long-running software engineering tasks where users can hand off difficult work with confidence; substantially improved vision supporting images up to 2,576 pixels on the long edge (~3.75 megapixels); better adherence to instructions; and more tasteful, creative outputs for professional artifacts like interfaces, slides, and documents. Pricing stays the same as Opus 4.6 at $5 per million input tokens and $25 per million output tokens. The model ships with new cyber safeguards that automatically detect and block high-risk cybersecurity requests, as the first step in testing mitigations before broader Mythos-class releases. It is available across Claude products, the API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.

Software Engineering and Autonomy Improvements

Opus 4.7 shows particular strength on the most difficult coding and engineering tasks. Early testers report the ability to delegate hard work that previously required close supervision.

Key reported advances:

Handles complex, long-running tasks with rigor, consistency, and self-verification of outputs.
Pays precise attention to instructions, reducing loose interpretations seen in prior models.
Users note it catches its own logical faults during planning, accelerates execution, and pushes through tool failures or hard problems.

Examples from testers:

On a 93-task coding benchmark, +13% resolution over Opus 4.6, solving four tasks neither prior Opus nor Sonnet 4.6 could.
On CursorBench, 70% vs. 58% for Opus 4.6.
On Rakuten-SWE-Bench, resolves 3x more production tasks with gains in code and test quality.
On Replit and other platforms, stronger performance on logs/traces, bug finding, and async workflows with fewer corrections.

Enhanced Vision and Multimodal Capabilities

The model supports significantly higher resolution images: up to 2,576 pixels on the long edge, more than three times prior Claude models (~3.75 megapixels).

This enables new use cases in fine visual detail work such as:

Computer-use agents reading dense screenshots.
Data extraction from complex technical diagrams.
Interpreting chemical structures or other precise visuals.

Testers highlight improvements in multimodal understanding for life sciences patents, dashboards, data-rich interfaces, and visual-acuity benchmarks (e.g., 98.5% vs. 54.5% in one penetration testing eval).

Safety Positioning and Cyber Safeguards

Opus 4.7 is positioned as less broadly capable than the limited Claude Mythos Preview but with targeted safeguards. Following Project Glasswing (which highlighted AI cyber risks and benefits), Anthropic is using this model to test mitigations first.

New features:

Automatic detection and blocking of requests indicating prohibited or high-risk cybersecurity uses.
Cyber Verification Program for legitimate security professionals (vulnerability research, penetration testing, red-teaming).

This approach allows learning from real-world deployment before any broader Mythos release.

Pricing, Availability, and Platform Support

Pricing is unchanged from Opus 4.6: $5/M input tokens, $25/M output tokens.

Availability at launch:

All Claude products (Pro, Max, Team, Enterprise).
Claude API (use claude-opus-4-7).
Amazon Bedrock.
Google Cloud Vertex AI.
Microsoft Foundry.
Integrations including Claude Code, Cursor, GitHub Copilot.

The model includes API breaking changes versus Opus 4.6; migration guidance is provided.

Early Tester Feedback

Testers across companies (Hex, Replit, Solve Intelligence, Cursor, Notion, Quantium, Vercel, CodeRabbit, Genspark, Warp, Bolt, Factory, XBOW, Qodo, Harvey, Databricks, Ramp, and others) report consistent themes:

Stronger autonomy and long-horizon task completion.
Better instruction following and honesty about limits.
Improved design taste and creative reasoning for professional outputs.
Efficiency gains (same or better quality at lower effective cost in some cases).
Specific benchmark lifts in coding, agentic workflows, document reasoning, and finance/legal tasks.

One example: autonomously built a complete Rust text-to-speech engine (model, kernels, demo) and verified it.

Why this story matters

The Claude Opus 4.7 release on April 16, 2026, illustrates Anthropic’s dual-track strategy: advancing generally available models for broad use in software engineering and professional work while carefully gating the most powerful capabilities (Mythos Preview) behind additional safeguards. Concrete improvements in autonomy, vision resolution, instruction fidelity, and self-verification—paired with unchanged pricing and wide platform availability—provide measurable progress for developers and enterprises. The inclusion of production cyber safeguards on this model marks an early, deliberate step toward responsible scaling of high-capability systems. User reports and benchmarks quantify the lift over Opus 4.6, highlighting reliability gains critical for agentic and long-running workflows.

Sources

Anthropic: “Introducing Claude Opus 4.7” (April 16, 2026). https://www.anthropic.com/news/claude-opus-4-7
Anthropic Claude Opus 4.7 System Card and related documentation.
Cross-referenced platform announcements (AWS Bedrock, Google Cloud Vertex AI, Microsoft Foundry) and early tester reports confirming capabilities, pricing, and availability at launch.

Featured Image Alt Text

Abstract visualization of Claude Opus 4.7 model release on April 16, 2026, showing layered improvements in coding autonomy, high-resolution vision for diagrams and screenshots, and cyber safeguard shields, with Anthropic logo and benchmark upward arrows