source: techcrunch ai: anthropic launches claude sonnet 5 as a cheaper way to run agents
level: technical
anthropic released claude sonnet 5, a new midsize model focused on agentic tasks like planning, tool use, and autonomous operation. it can use browsers and terminals, and runs independently at a level that previously needed larger, more expensive models. the model is now the default for free and pro plans, with pricing at $2 per million input tokens and $10 per million output tokens until august 31, then $3 and $10. this makes it cheaper than opus 4.8, gpt-5.5, and gemini 3.1 pro, but more expensive than gemini 3.5 flash.
sonnet 5 shows gains over its predecessor sonnet 4.6 on reasoning, tool use, coding, and knowledge work. on an agentic coding benchmark, it scored 63.2%, compared to opus 4.8's 69.2% and sonnet 4.6's 58.1%. on a knowledge work benchmark, it slightly outperformed opus 4.8. testers noted it completes complex tasks that earlier versions would stall on, and it checks its own output without being asked. for example, a zapier engineer said it finished a two-part job updating salesforce and sending a launch announcement end to end.
safety evaluations show sonnet 5 has lower rates of undesirable behaviors like cooperation with misuse and deception than sonnet 4.6. it is better at refusing malicious requests and avoiding prompt-injection attacks, and it hallucinates less. however, it is not as safe as opus 4.8 or claude mythos preview on misaligned behavior. lovable's co-founder said it refuses unsafe requests cleanly and consistently, which matters when giving powerful tools to many builders.
why it matters: cheaper agentic models let developers build autonomous ai features at lower cost, making agentic workflows more accessible.
source: techcrunch ai: anthropic launches claude sonnet 5 as a cheaper way to run agents