What's New in Claude Opus 5?
Anthropic has officially launched Claude Opus 5, the latest evolution in the Opus model family. Engineered for high-stakes software engineering, long-running agent workflows, and deep knowledge work, Opus 5 delivers near-frontier intelligence at half the operating cost of previous top-tier models.
Whether you are automating complex business logic, conducting scientific research, or deploying autonomous software agents, Opus 5 delivers greater accuracy and thoroughness while generating fewer tokens.
Key Capabilities & Breakthroughs
New Benchmark Standard for Coding & Knowledge Work
Surpasses previous models on Frontier-Bench v0.1 and GDPval-AA, establishing new industry baselines for developer and analyst tooling.
Scores within 0.5% of Claude Fable 5 on CursorBench 3.2 at max effort, but at 50% lower cost per task.
On ARC-AGI 3, Opus 5 achieved 3x higher performance than the next-best model in solving novel reasoning problems.
Unmatched Agentic Thoroughness
On OSWorld 2.0 (computer use) and Zapier AutomationBench, Opus 5 outpaces competing models, achieving 1.5x the pass rate for the same cost per task.
In early testing, when given a raster drawing of a machine part without direct visual access, Opus 5 written its own computer vision pipeline to extract the geometry from raw pixels and successfully reconstructed a full 3D FreeCAD model.
Domain-Specific Mastery
Software Engineering: Catches surface bugs and resolves underlying edge cases that competing models miss entirely.
Life Sciences: Outperforms Opus 4.8 across all life science evaluations, including a 10.2 percentage point jump in organic chemistry spectroscopy tasks.
Legal & Financial Analysis: Enterprise testing by Box showed an 11% boost in data analysis workflows and a 17% improvement in due diligence tasks.
Engineered for Speed, Control, & Safety
Features granular effort settings, allowing teams to tune reasoning levels to generate up to 26% fewer tokens on average.
Includes a Fast mode operating at 2.5x the default speed.
Rated as Anthropic's most aligned model to date (audit score of 2.3), with significantly lower rates of deceptive behavior.
Performance & Cost Comparison
Metric / Feature | Claude Opus 5 | Claude Opus 4.8 |
Input Price / M Tokens | $5.00 | $15.00 |
Output Price / M Tokens | $25.00 | $75.00 |
Frontier-Bench Coding | New State-of-the-Art | Baseline |
ARC-AGI 3 (Novel Logic) | 3x Next-Best Model | Baseline |
Token Efficiency | ~26% fewer tokens generated | Standard token output |
How to Get Started
Claude Web & Desktop Apps: Opus 5 is live today as the default model on Claude Max and the strongest model tier on Claude Pro, Team, and Enterprise.
Developers & API Users: Available via the Claude API (
claude-opus-5), as well as Amazon Web Services (Bedrock), Google Cloud, and Microsoft Foundry.Cost Savings: Save up to 90% with prompt caching and 50% with batch processing.
