The model became available across Anthropic's consumer app, developer platform and enterprise tools, the company said. It will serve as the default model for subscribers to Claude Max, Anthropic's higher-tier consumer plan, and will be the most capable option offered to Claude Pro subscribers.
Opus 5 is priced at $5 per million input tokens and $25 per million output tokens, the same rate Anthropic charged for its previous flagship, Opus 4.8. Tokens are the chunks of text AI models use to process language, and per-token pricing is the standard way developers pay to build software on top of models like Claude.
Anthropic said Opus 5 topped rival models, including its own predecessor, on a range of internal and third-party benchmarks measuring software engineering, business-automation tasks and general problem-solving. On one evaluation involving novel puzzle-solving, the company said Opus 5 scored roughly three times higher than the next-best model it tested. On a computer-use benchmark, it said Opus 5 outperformed every competing model regardless of cost, and beat its own top-tier model's best result at a fraction of the price.
The company also pointed to gains in scientific research applications, saying Opus 5 outperformed its predecessor on evaluations covering structural biology, organic chemistry and bioinformatics.
Anthropic has been competing against OpenAI, Google and other AI developers to release increasingly capable models while managing the high computing costs that come with training and running them. The company has increasingly promoted adjustable "effort" settings that let customers dial performance up or down to manage expenses, a feature it emphasized again in Friday's release.
Anthropic said Opus 5 remains behind Claude Mythos 5, a more tightly restricted model the company offers to a limited set of partners, on cybersecurity and biological-research tasks that carry higher misuse risks. It said it had eased some safety restrictions on Opus 5 for source-code vulnerability searches while maintaining tighter controls on exploit development and other higher-risk security work.
The company said an internal safety review found Opus 5 to be its most compliant model to date with Anthropic's published AI-safety principles, showing lower rates of deceptive behavior than earlier models and a reduced susceptibility to being misused.
Anthropic, founded in 2021 by former OpenAI researchers, has raised billions of dollars from investors as it competes for a share of the fast-growing generative AI market.
I truly appreciate you spending your valuable time here. To help make this blog the best it can be, I would love your feedback on this post. Let me know in the comments: How could this article be better? Was it clear? Did it have the right amount of detail? Did you notice any errors?
If you found any of the articles helpful, please consider sharing it.