Anthropic has released Claude Opus 5, the newest addition to its Opus model family. The new AI model is designed for knowledge work, scientific research, computer reasoning, and software engineering tasks. Positioned as Anthropic’s default model for Claude Max and the strongest model for Claude Pro, Opus 5 delivers near-frontier intelligence while operating at half the cost of Claude Fable 5.
One of the noteworthy highlights of Claude Opus 5 is its outstanding performance for the same cost as its predecessor, Opus 4.8. It establishes a new state-of-the-art software engineering task and outperforms previous Opus models.
Claude Opus 5 Excels Across Benchmarks
- On Frontier-Bench v0.1, Opus 5 outperforms all other models and doubles Opus 4.8’s performance at a lower cost per task.
- On CursorBench 3.2, the model performs within 0.5% of Claude Fable’s peak score while costing half as much per task. The model also delivers outstanding results on knowledge work and problem-solving tasks:
- The model excels on ARC-AGI 3, where it scores three times as high as the next-best model.
- Additionally, Claude Opus 5 achieves outstanding results on Zapier AutomationBench and OSWorld 2.0, demonstrating strong capabilities in business tasks and computer use.
Beyond software engineering tasks, Opus 5 shows significant improvements over Opus 4.8 for scientific research, mainly in the life sciences, organic chemistry, structural biology, and bioinformatics. Anthropic claims measurable gains over Opus 4.8, including higher performance in molecular structure analysis and protein function prediction, making the model ideal for advanced research applications.
Real-World Problem Solving
Anthropic has also confirmed that Claude Opus 5 is stronger at verifying its work and making changes until it succeeds. In evaluations and early testing, the tech giant found the following:
- Opus 5 created its own computer vision pipeline to reconstruct a 3D FreeCAD model from raw pixels, then used to reconstruct the full machine part.
- The model found the root cause of the software bug and fixed an edge case that the community patch had missed.
- It built its own test harness to verify that the code correctly interpreted the exchange data.
These examples highlight its improved agentic capabilities and reliability in real-world engineering workflows.
Safety and Alignment
Safety remains a key focus of the release. Claude Opus 5 is Anthropic’s most aligned model to date, exhibiting lower deceptive behavior, being less prone to misuse, and showing stronger adherence to its AI safety principles than previous Opus, Sonnet, and Fable models. It avoids risky actions while delivering reliable performance for enterprise and research use.
To support responsible use, Opus 5 includes safeguards that allow for defensive cybersecurity work and scientific research. It restricts high-risk activities such as exploit generation, penetration testing, and binary-based vulnerability scanning.
According to Anthropic, “Opus 5 cyber classifiers are proportionally less restrictive than those on Fable 5.” At the same time, approved organizations can access fewer restrictions through Anthropic’s Cyber Verification Program (CVP), which facilitates cybersecurity work and could hinder the model’s safeguards.
Availability, Pricing, and How to Get Started?
Claude Opus 5 is now available on all platforms, including the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry. It is priced at $5 per million input tokens and $25 per million output tokens, which is the same as Opus 4.8.
Developers can get started with Opus 5 through the Claude API. Anthropic also introduced a Fast mode that runs approximately 2.5 times faster, along with two updates in beta, including mid-conversation tool changes on the Claude Platform and Automatic Fallbacks on the API.
As Anthropic continues to push the boundaries of frontier AI, Claude Opus 5 marks another step toward faster, safer, and more capable models for developers, enterprises, and researchers alike.
Curious to dive into the latest happenings around the tech realm? Don’t miss out on visiting our website.
Our Latest Blogs:
What is AI Observability and Why Does it Matter for Production Machine Learning Systems?
Vulnerability Scanning vs Penetration Testing: What’s the Difference