Claude Opus 5
Introducing Claude Opus 5
Date: July 24, 2026
Developer: Anthropic
Anthropic is proud to announce the release of Claude Opus 5, a model characterized by its proactive nature and thoughtful approach to problem-solving. It delivers intelligence that rivals the frontier capabilities of Claude Fable 5, yet it is available at the same price half the cost.
🚀 Positioning & Availability
Opus 5 is engineered for daily integration, prioritizing efficiency without sacrificing power.
- Claude Max: Now the default model.
- Claude Pro: The most powerful option available.
While Opus 5 has established a new state-of-the-art (SOTA) on GDPval-AA and Frontier-Bench for knowledge work and coding, it is important to note that it still trails behind Mythos 5 in specialized cybersecurity tasks.
📈 Performance and Cost-Efficiency
Opus 5 represents a massive leap over its predecessor, Opus 4.8, offering superior performance for the same expenditure. A key feature is the Effort Setting, which allows users to balance intelligence against token consumption.
Benchmark Highlights
The following table summarizes the model's dominance across key evaluations:
| Benchmark | Performance Detail | Cost Comparison |
|---|---|---|
| Frontier-Bench v0.1 | Surpasses all other models | performance of Opus 4.8 |
| CursorBench 3.2 | Within of Fable 5 (at max effort) | of the cost per task |
| ARC-AGI 3 | higher score than the next-best model | N/A |
| Zapier AutomationBench | pass rate of next-best model | Equal cost per task |
| OSWorld 2.0 | Outperforms all models at any cost | Beats Fable 5 at cost |
Other notable victories include:
HLEDeepSearchQAGDPval-AA v2
🧪 Scientific Research & Visuals
Opus 5 is a significant upgrade for the scientific community, particularly in the life sciences (bioinformatics, organic chemistry, and structural biology).
Quantitative Improvements over Opus 4.8:
- Organic Chemistry: (specifically in deriving molecular structures from spectroscopy data).
- Protein Analysis: (predicting functional changes based on sequence variations).
Visual Capabilities
The model can now generate sophisticated visual outputs, such as:
- Aerodynamics: Visualizing air flow over various objects in a wind tunnel.
- Biology: Creating interactive, simplified illustrations of cellular structures.
🛠️ Agency and Iterative Problem Solving
Opus 5 excels at self-verification and persistent iteration. Rather than giving up, it develops its own tools to overcome obstacles.
Case Study 1: The FreeCAD Challenge When tasked with rebuilding a machine part from a drawing—but intentionally blocked from viewing the image—Opus 5 did not fail. Instead, it wrote its own computer vision pipeline to extract geometry from raw pixels and successfully reconstructed the 3D model. No other model succeeded in five attempts.
Case Study 2: The Open-Source Bug Faced with a bug in a popular package manager, Opus 5 identified the root cause and fixed an edge case that the community's official patch had overlooked. Competitors merely treated the symptoms.
Case Study 3: The Trading Feed An engineer used Opus 5 to build a market data feed for a new exchange in one session. Because there was no live feed for validation, the model built its own test harness to ensure the data parsing was accurate.
💬 Early Access Feedback
Users across various industries have reported transformative results:
- FrontierCode 1.1: Near-Fable performance at cost.
- Devin: Exceptional at root-cause analysis and complex debugging.
- Zapier: Topped the AutomationBench; successfully executed a full churn-prevention workflow (Flagging Alerting Summarizing).
- Genomics: Acts as a "careful scientist," utilizing independent cross-checks and rigorous statistical tests to eliminate confounders.
- Lovable: The biggest jump since version 4.5, specifically in 3D work, games, and animations.
Internal Evaluation Note: Opus 5 is not only more powerful—it is more stable. On agentic coding tasks, it is 22% better than Opus 4.7 and exhibits significantly lower variance between runs.
