Anthropic Claude launches Sonnet 5 as a cheaper way to manage clients

As freight forwarder capabilities become stakes among established model companies, Anthropic is launching the Claude Sonet 5, a more powerful, agent-oriented version of the mid-sized laboratory model.

“It can make plans, use tools like browsers and terminals, and operate autonomously at a level that, just a few months ago, would have required larger, more expensive models,” Anthropic said in a report. Blog post.

This framework mirrors what OpenAI and Google have said in their recent releases. OpenAI’s GPT-5.6 Sol Launched in preview last week, it’s also the company’s most agented model yet, allowing users to split work across sub-agents for longer independent tasks. Google Gemini 3.5 FlashLaunched in May, it was billed as a transformation from a conversational chatbot to a powerful tool that plans, builds, and iterates on real work with minimal human input.

The Sonnet 5 offering is confirmation that agent capacity is the new fundamental expectation at every price level. The difference now will not be who can do the agents’ work better, but how cheaply it can do it and how reliable it can be without human supervision.

Sonnet 5 promises near-perfect performance Opus 4.8but at much lower costs. Starting Tuesday, Claude Sonnet 5 will be the default model for the Free and Pro plans, available for every subscription.

At launch, Sonnet 5 is priced at $2 per million input tokens and $10 per million output tokens until August 31, after which the price will jump to $3 per million input tokens and $15 per million output tokens. This makes Sonnet 5 cheaper than Opus 4.8, as well as OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro. (It’s still more expensive than the Gemini 3.5 Flash.)

The new model also shows significant improvements over its predecessor Sonnet 4.6, Released in Februaryon effective performance such as thinking, tool use, software coding, and cognitive work, according to Anthropic.

For example, in one benchmark, Sonnet 5 scored 63.2% in proxy encryption, compared to Opus 4.8’s 69.2% and Sonnet 4.6’s 58.1%. In the cognitive work benchmark, the Sonnet 5 actually slightly outperforms the Opus 4.8, which is known for winning at the toughest problems like making accurate judgments and deep searching.

“Opus 4.8 is still the model of choice for higher resolution in these tasks, but Sonnet 5 provides developers with lower-priced, much higher-quality options than were previously available,” Anthropic says. “Between Sonnet 5 and Opus 4.8, users can adjust the effort level to find the right balance between cost and performance.”

According to testers mentioned in the blog post, the Sonnet 5 also excels at finishing complex tasks where previous versions of the model would have stopped working and “check their own output without being explicitly asked to do so.”

“We handed Claude Sonnet 5 a two-part task — updating Salesforce account levels, and sending the launch announcement to enterprise contacts — and it was done from start to finish,” Daniel Shepherd, a senior engineer at Zapier, said in a statement. “It used to stop halfway through. As for day-to-day automation, it’s a no-brainer.”

In terms of safety, Sonnet 5 also shows a lower rate of “unwanted behaviors” such as cooperation with abuse and phishing compared to its predecessor, making it safer for use in proxy contexts. It is best to reject malicious requests and avoid hijacking attempts in spot injection attacks. He also hallucinates and engages in sycophantic behavior less frequently than in Sonnet 4.6.

However, it’s not on the same level as Opus 4.8 and Claude Mythos Preview when it comes to deviant behavior. “Evaluations also show that it has a significantly lower ability to perform critical cybersecurity tasks than our current Opus models,” the blog post said.

Claude Sonet 5 “rejects unsafe requests cleanly and consistently,” beloved co-founder Fabien Hedin said in a statement.

“At Lovable, we put powerful tools in the hands of millions of construction workers,” Hedden said. “A model that knows when to say no is just as important as a model that knows how to build.”

Updated to correct that the price of Director Tokens is $15 per million Director Tokens after August 31st.