Anthropic has launched Claude Opus 5, its latest artificial intelligence model, claiming it delivers significant improvements in coding, research and business tasks while approaching the performance of Claude Fable 5 on several benchmarks at a lower cost.
The company said Claude Opus 5 succeeds Claude Opus 4.8 and is available immediately across Claude's apps and API. It is now the default model for Claude Max subscribers and the most advanced model included with the Claude Pro plan.
According to Anthropic, Opus 5 achieved results within 0.5 percentage points of Claude Fable 5 on the CursorBench 3.2 software engineering benchmark while costing roughly half as much per task.
The benchmark evaluates AI coding agents on real-world software development challenges.
Anthropic also said Opus 5 outperformed Fable 5 on the OSWorld 2.0 benchmark, which measures how effectively AI models operate computers and complete tasks across applications and files, while using just over one-third of the cost.
The launch comes shortly after Claude Fable 5 moved behind a pay-as-you-go credit system, potentially making Opus 5 a more affordable option for Pro and Team subscribers.
Improved software engineering performance
Anthropic said Claude Opus 5 more than doubled the performance of Claude Opus 4.8 on Frontier-Bench, a benchmark designed to test AI agents on complex software engineering tasks.
The company said the model is better at identifying bugs, checking its own work and continuing difficult programming tasks instead of stopping after a partial solution.
In one internal test, Opus 5 reportedly identified an edge case that had been missed by an existing community-developed software patch.
Anthropic noted that benchmark results were based on its own internal testing and that real-world performance may vary depending on workloads.
Despite the performance gains, API pricing remains unchanged at $5 per million input tokens and $25 per million output tokens, matching the cost of Claude Opus 4.8.
