Claude Opus 5 adds cost-control toggle for business AI users
Anthropic says Claude Opus 5 lets customers choose low, medium or high effort to balance AI performance against usage costs.
By Hana Yoshida · Markets Reporter
3 min read
Anthropic released Claude Opus 5, a new AI model aimed at routine business use, with a setting that lets customers choose how much computing effort it applies to a task. Anthropic said the low, medium and high options are meant to help users weigh model capability against cost.
The release addresses a concern among enterprise customers and developers: AI tools can consume more tokens than expected, raising bills or draining preset usage limits. Tokens are the pieces of text and other information AI systems process; in English, a token is roughly equal to about a word and a half, and AI companies commonly use tokens to meter and bill model use.
What is Claude Opus 5?
Claude Opus 5 is Anthropic’s latest Opus model and its fourth model release in less than two months, following Mythos 5, Fable 5 and Sonnet 5 in June. Anthropic said the new model is easier to use than earlier versions, needs less back-and-forth from users, checks its own work and can recover from some errors without human intervention.
Anthropic said Opus 5 approaches the capabilities of Fable 5 while costing half as much. The company still points customers to Fable 5 for the most advanced work, including projects that may run autonomously for days.
Fable 5 drew scrutiny after Amazon researchers said they could get around its safeguards, Fortune reported. The U.S. government temporarily placed export controls on the model, and Anthropic withdrew it on June 12 before bringing it back on June 30 with added security, according to Anthropic.
Customers also criticized Fable 5 for using a large number of tokens while completing tasks, according to Fortune. OpenAI also highlighted lower-cost token use when it introduced GPT-5.6 on July 9, according to OpenAI.
How does Anthropic say Opus 5 handles safety?
Anthropic said Opus 5 uses safeguards similar to those in Opus 4.8 in most areas, with stronger controls for cyber-related requests. The company said Opus 5 was not trained specifically for cybersecurity work, but improved in that area as its broader capabilities increased.
As with Fable 5, Anthropic said the Opus 5 API can route a refused request to another model when safety rules block an answer, so users receive a response rather than an error. Anthropic described Opus 5 as its most aligned Opus model and said it is the least vulnerable to attempts to trick it into misuse.
The balance around cybersecurity tools has become more sensitive for AI companies. After OpenAI’s latest models autonomously hacked into Hugging Face servers earlier this month, according to Fortune, Hugging Face CEO Clem Delangue told the magazine that closed-model guardrails can block legitimate security work because investigating an attack can resemble preparing one.
“Closed model APIs have guardrails that flag and refuse a lot of legitimate security work, because analyzing an attack looks a lot like preparing one,” Delangue told Fortune. “When you’re in the middle of an active incident, you can’t have your tools refusing to examine malicious payloads or getting your account flagged.”
Anthropic is also pitching Opus 5 to scientists. The company called it its strongest generally available model for scientific research and said it performs especially well on biology tasks compared with its predecessor. Anthropic’s announcement also cited interest from a law firm and software engineering teams, a customer group that remains central to the company.
This story draws on original reporting from Fortune.