Anthropic has introduced another new AI model, which is already available to users — Claude Opus 5. It is called a «thoughtful and proactive model» that «approaches the advanced intelligence» of Claude Fable 5, but costs half the price.
Anthropic has introduced another new AI model, which is already available to users — Claude Opus 5. It is called a «thoughtful and proactive model» that «approaches the advanced intelligence» of Claude Fable 5, but costs half the price.
When it comes to coding and data processing, Opus 5 performs even better than other top Anthropic AI models on the Frontier-Bench and GDPval-AA benchmarks, although it still lags behind Mythos 5 on cybersecurity tasks.
The company points out that Claude Opus 5 provides significantly improved performance at the same price as its predecessor, Opus 4.8.
In particular, Opus 5 excels at software development tasks.
For example, in Frontier-Bench v0.1, the new AI model outperforms all other models and more than doubles Opus 4.8. At the same time, it requires less overhead to complete tasks. In CursorBench 3.2, the model shows results within 0,5% of the peak score of Fable 5.
Meanwhile, in the ARC-AGI 3 assessment, Opus 5 scored three times higher than the next best model. On Zapier AutomationBench, which measures whether models can complete business tasks from start to finish, Opus 5’s success rate is about 1.5 times higher than the next best model on the list.
«Even at the lowest effort settings, the Opus 5 performs more tasks than any other model. In OSWorld 2.0, the Opus 5 outperforms every other model regardless of price, beating the best result of the Fable 5 at just over a third of the cost,» the Anthropic blog notes .
It is also the company’s best and most economical model according to several related assessments:
Opus 5 performs significantly better than Opus 4.8 for scientific research, particularly in structural biology, organic chemistry, and bioinformatics.
The AI company also demonstrated how the model copes with creating complex visual effects.
In terms of cybersecurity, Opus 5 finds vulnerabilities in source code, but at the same time blocks binary-based vulnerability scanning (a method more often associated with attackers), penetration testing, and exploit creation.
Claude Opus 5 is available today on all platforms at a price of $5 per million input tokens and $25 per million output tokens. Developers can get started with claude-opus-5 on the Claude API.
Recall that in early July, Anthropic introduced the «most agent model» of all existing in this line — Claude Sonnet 5.
