Claude Opus 5 Release Notes

Anthropic has released Claude Opus 5, a model designed for high-efficiency daily use that provides near-frontier intelligence (approaching Claude Fable 5) at half the cost per task. It is now the default model on Claude Max and the strongest model available on Claude Pro.

State-of-the-Art Performance in Coding and Knowledge Work

Claude Opus 5 establishes new state-of-the-art results on several key evaluations for software engineering and problem-solving, often outperforming predecessors and competing models at a lower cost.

Software Engineering Benchmarks

  • Frontier-Bench v0.1: Opus 5 surpasses all other models and more than doubles the performance of Opus 4.8 while reducing the cost per task.
  • CursorBench 3.2: At maximum effort, Opus 5 performs within 0.5% of Fable 5's peak score at half the cost per task. It outperforms all other models at high, xhigh, and max effort levels for a given cost.
  • OSWorld 2.0: Opus 5 outperforms every other model at any given cost, exceeding Fable 5's best result at approximately one-third of the cost.

Knowledge Work and Problem Solving

  • ARC-AGI 3: Opus 5's score is three times higher than the next-best model on this novel problem-solving evaluation.
  • Zapier AutomationBench: Opus 5 achieves a pass rate approximately 1.5× the next-best model for the same cost per task. Even at its lowest effort setting, it passes more tasks than any other model.
  • Frontier-Bench and GDPval-AA: Opus 5 is the new state-of-the-art on these coding and knowledge work evaluations.

Scientific Research and Visual Capabilities

Opus 5 introduces significant improvements in scientific research, particularly in the life sciences, and enhanced visual output capabilities.

Life Sciences Improvements

Opus 5 outperforms Opus 4.8 across all life sciences evaluations, including bioinformatics, structural biology, and organic chemistry. Key gains include:

  • Organic Chemistry: A 10.2 percentage point increase over Opus 4.8 in inferring molecular structures from spectroscopy data.
  • Protein Research: A 7.7 percentage point increase in predicting how protein sequence variations affect function.

Agentic Behavior and Practical Application

Claude Opus 5 demonstrates increased agency, thoroughness, and the ability to verify its own work through iteration.

Technical Case Studies

  • Autonomous Tooling: In a Frontier-Bench task, Opus 5 wrote its own computer vision pipeline to extract geometry from raw pixels to rebuild a 3D FreeCAD model when it was denied direct viewing capabilities.
  • Bug Resolution: Opus 5 identified the root cause of a real bug in a popular open-source package manager and fixed an edge case that previous community patches had missed.
  • Infrastructure Development: A trading firm engineer used Opus 5 to build a market data feed in a single session, including the creation of its own test harness for validation.

Industry Feedback

Early-access partners have highlighted specific strengths:

  • Devin: Noted strength in difficult debugging and root-cause analysis.
  • Lovable: Reported a 22% improvement over Opus 4.7 on hardest agentic coding tasks with lower variance.
  • Box: Found an 8% overall performance increase over Opus 4.8, with 11% gains in data analysis and 17% in due diligence workflows.
  • Legal Agents: Observed similar performance to max reasoning levels while generating 26% fewer tokens than Opus 4.8.

Alignment, Safety, and Cybersecurity

Opus 5 is described as Anthropic's most aligned model to date, adhering more closely to Claude's Constitution than Opus 4.8, Sonnet 5, or Fable 5.

Behavioral Audit and Risk Mitigation

  • Alignment Score: Opus 5 scored 2.3 on overall misaligned behavior in automated behavioral audits, the lowest among recent models.
  • Cybersecurity Limitations: While Opus 5 is close to Mythos 5 at identifying vulnerabilities, it remains substantially behind Mythos 5 in the exploitation of those vulnerabilities (as shown on the OSS-Fuzz benchmark).
  • Biology Research: Opus 5 is the most capable generally available model for scientific research but remains behind Mythos 5 in long-running, autonomous research tasks.

Safeguards and Classifiers

  • Cyber Classifiers: These are less restrictive than Fable 5's, allowing vulnerability discovery in source code but blocking binary-based scanning, penetration testing, and exploit generation. These classifiers are expected to intervene 85% less often than those on Fable 5.
  • Fallbacks: Requests flagged by safety classifiers on Opus 5 or Fable 5 can now automatically route to another model via the API.

Pricing and Availability

Claude Opus 5 is available on all platforms with the following pricing:

  • Input Tokens: $5 per million tokens.
  • Output Tokens: $25 per million tokens.
  • Fast Mode: Available at twice the base price, providing approximately 2.5× the default speed.

Beta Feature Releases

  • Mid-conversation tool changes: Developers can now modify available tools within a conversation without invalidating the prompt cache.
  • Automatic fallbacks: API users can route flagged requests to a different model to avoid being blocked.

Sources

Related