Moving the artificial intelligence (AI) arms race from conversational chatbots to autonomous agents, Anthropic on Tuesday released Claude Sonnet 5.
The new midsize model brings advanced, self-directed capabilities to everyday professionals and developers at a fraction of the cost of previous systems, while simultaneously lowering critical cybersecurity risks.
Welcome to the latest turn in the AI industry’s shift from systems that merely answer queries toward agentic AI. Anthropic’s latest release signals that these autonomous capabilities are becoming standard across all pricing tiers, forcing top labs to compete heavily on cost and reliability.
According to Anthropic, Sonnet 5 closes the performance gap with the company’s premium Opus 4.8 model, yet commands a much lower price point. At launch, Sonnet 5 is priced at $2 per million input tokens and $10 per million output tokens through Aug. 31, undercutting rivals like OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro. After the promotional period, pricing will adjust to $3 per million input and $15 per million output tokens. To ensure widespread adoption, Anthropic has made Sonnet 5 the new default model for all Free and Pro subscription plans.
The upgrades over its predecessor, Sonnet 4.6, are substantial.
On the Terminal-bench 2.1 agentic coding benchmark, Sonnet 5 achieved an 80.5% success rate compared to Sonnet 4.6’s 67%. In separate evaluations, Sonnet 5 scored 63.2% on agentic coding, trailing just behind Opus 4.8’s 69.2%.
Notably, the new model slightly outperformed Opus 4.8 on specific knowledge-work metrics, offering users a highly efficient alternative for multi-step enterprise automation. Early testers reported that Sonnet 5 routinely checks its own output and successfully completes complex, multi-part workflows that previously stalled mid-task.
Beyond raw capability, Sonnet 5 addresses the safety liabilities inherent to autonomous software. The model features a lower rate of “undesirable behaviors” than Sonnet 4.6, demonstrating increased resilience against prompt-injection attacks and malicious requests. Crucially, Anthropic noted that Sonnet 5 poses significantly less cyber-risk than its flagship Opus and Mythos lines, rendering it safer for deployment in sensitive everyday environments.
The debut of Sonnet 5 comes amid intensifying industry competition and regulatory pressure.
Google recently deployed its 24/7 personal assistant, Gemini Spark, while Anthropic’s own Fable 5 was only recently re-released following U.S. government restrictions.
With OpenAI’s GPT-5.6 still under federal review, Sonnet 5 establishes a crucial market foothold. The launch underscores a broader reality for the AI sector: future dominance will not belong to the chatbot with the cleverest answers, but to the affordable, autonomous assistant that can reliably finish the job.