Anthropic has launched Claude Opus 5, the newest version of its advanced AI model built to handle coding, research, business operations, and other demanding tasks.
The company claims that Opus 5 delivers a significant performance improvement over Claude Opus 4.8, while maintaining the same API pricing.
Additionally, Anthropic suggests that Opus 5 approaches Claude Fable 5’s performance in certain coding and computer-related tests while being much more cost-effective per task.
Opus 5 is accessible across all Claude applications and its API. It is now the default model for Claude Max subscribers and represents the strongest option available with Claude Pro.
In testing conducted on CursorBench 3.2, which evaluates coding agents performing real software development tasks, Opus 5 nearly matched Fable 5’s highest score, finishing within 0.5% at maximum effort. Moreover, Anthropic states that Opus 5 achieved this result at half the cost per task.
Opus 5 also outperformed Fable 5’s best score on OSWorld 2.0 while costing just over a third of the price. This benchmark assesses how well AI agents can operate computers and complete tasks across various applications and files.
For Pro and standard Team subscribers, Opus 5 could fill the gap left by Fable 5, which transitioned to a pay-as-you-go credit system on July 20.
READ MORE: China AI industry recruits teenagers to solve massive AI talent shortage
Anthropic claims that Opus 5 more than doubled the performance of Opus 4.8 on Frontier-Bench, a benchmark that tests AI agents on challenging software engineering tasks.
It also completed these tasks at a lower average cost.
The company says Opus 5 is better at checking its own work, finding bugs, and sticking with tough problems rather than stopping after a quick solution.
In one example, Opus 5 found an edge case that a previous community patch had missed.
These results are based on Anthropic’s own testing, meaning that actual performance may vary across different workloads.
The pricing for Claude Opus 5 remains unchanged from Opus 4.8, costing $5 per million input tokens and $25 per million output tokens.