Anthropic’s Claude Sonnet 5 pushes cheaper agentic models

Anthropic is rolling out Claude Sonnet 5 as a more powerful, agentic version of its midsize model—aimed at getting autonomous planning and tool use into the hands of more developers without the higher price tag. Starting Tuesday, Sonnet 5 becomes the default f

On Tuesday. Anthropic is asking developers to stop thinking of agentic AI as a luxury reserved for the biggest. most expensive models. Claude Sonnet 5 is being positioned as a midsize alternative that can plan. use tools like browsers and terminals. and run autonomously at a level that Anthropic says used to require larger and more expensive systems.

In a blog post, the company described Sonnet 5’s core pitch plainly: it can make plans, use tools, and run on its own. The statement echoes the framing from other major labs as they race to treat autonomous capability as a baseline feature, not a differentiator.

OpenAI’s GPT-5.6 Sol. launched in preview last week. is being presented as its most agentic model yet. with the ability to split work across subagents for longer autonomous tasks. Google’s Gemini 3.5 Flash. launched in May. was pitched as a shift from conversational chat to an agentic tool that plans. builds. and iterates on real work with minimal human input.

In that context, Anthropic’s message lands as a pricing-and-reliability challenge. The differentiator, Anthropic appears to be saying, won’t just be who can do agentic work best. It will be who can do it more cheaply and more reliably without needing constant human oversight.

Sonnet 5 will be the default model for free and Pro plans starting Tuesday. and it will be available for every subscription. At launch, the model is priced at $2 per million input tokens and $10 per million output tokens through August 31. After that, the price will increase to $3 per million input tokens and $10 per million output tokens.

Anthropic says that pricing makes Sonnet 5 cheaper than Opus 4.8, as well as OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro. Sonnet 5 is still more expensive than Gemini 3.5 Flash.

The company also points to performance benchmarks meant to show that the lower cost doesn’t mean cutting corners. Anthropic says Sonnet 5 promises performance close to Opus 4.8. but at much lower costs. and it frames those gains as significant improvements over its predecessor. Sonnet 4.6. released in February.

In agentic performance areas like reasoning, tool use, software coding, and knowledge work, Anthropic says Sonnet 5 shows notable movement. On one benchmark, Sonnet 5 scores 63.2% on agentic coding. Opus 4.8 scores 69.2%, while Sonnet 4.6 lands at 58.1%. On a knowledge work benchmark. Sonnet 5 slightly outperforms Opus 4.8. a model described as known for winning on the hardest problems. including subtle judgment calls and deep research.

Anthropic also offers a practical comparison for developers. It says Opus 4.8 remains the model of choice for higher accuracy on these tasks. but that Sonnet 5 gives developers lower-priced options with much higher quality than what was previously available. The company says that between Sonnet 5 and Opus 4.8. users can adjust the effort level to find the right balance of cost and performance.

There is another claim in the pitch that speaks directly to the day-to-day frustration people hit when models stop midstream. Anthropic cites testers who say Sonnet 5 excels at finishing complex tasks where previous model versions would have stopped short. It also says the model “checks its own output without explicitly being asked.”.

Zapier developer automation is one of the examples Anthropic highlights. Daniel Shepard. a senior engineer at Zapier. said in a statement that the team gave Claude Sonnet 5 a two-part job—update Salesforce account tiers and send a launch announcement to enterprise contacts—and that it finished the work end to end. Shepard added that the same kind of task used to stall halfway. and that for day-to-day automation it has become “a no-brainer.”.

Safety is part of the rollout too, especially given the agentic framing. Anthropic says Sonnet 5 shows a lower rate of “undesirable behaviors,” including cooperation with misuse and deception, than its predecessor. The company says it’s better at refusing malicious requests and sidestepping hijack attempts in prompt-injection attacks.

Anthropic also says Sonnet 5 hallucinates and engages in sycophantic behavior at a lower rate than Sonnet 4.6. Still, it draws a line between Sonnet 5 and its top-tier alternatives. The company says Sonnet 5 is not on the same level as Opus 4.8 and Claude Mythos Preview when it comes to misaligned behavior. “Evaluations also show that it has a much lower ability to perform dangerous cybersecurity tasks than our current Opus models. ” the blog post reads.

Lovable co-founder Fabian Hedin reinforced the safety theme in a statement. Hedin said that Claude Sonnet 5 “refuses unsafe requests cleanly and consistently. ” adding that Lovable is putting powerful tools in the hands of millions of builders and that “a model that knows when to say no is just as important as one that knows how to build.”.

The story underneath the pricing and benchmarks is about who gets access to autonomy—and at what cost. With Sonnet 5 becoming the default model for free and Pro plans and stepping into a lower price tier. Anthropic is betting that more developers will want to build agentic workflows now. not later. and that they’ll increasingly choose based on how far the model can go on its own.

Anthropic Claude Sonnet 5 agentic AI pricing tokens tool use Zapier safety prompt injection