Anthropic Launches Claude Sonnet 5 Model Offering Enhanced AI Agent Capabilities at Lower Costs

Anthropic introduced Claude Sonnet 5, a more powerful version in its Sonnet series, which serves as the mid-tier model of the Claude family. The new model features improved agent capabilities, including autonomous task execution, tool usage such as browsers and terminals, and programmatic planning, functions that previously required larger, more expensive models. Claude Sonnet 5 is now available to all users, including free-tier subscribers with limited usage quotas.

The model delivers performance close to Anthropic's advanced Opus 4.8 but at significantly reduced costs. To mark the launch, pricing is set at $2 per million input tokens and $10 per million output tokens until August 31, after which input token costs will rise to $3 per million. This pricing undercuts OpenAI's GPT-5.5 and Google's Gemini 3.1 Pro, although it remains more expensive than Gemini 3.5 Flash.

Compared to its predecessor Sonnet 4.6, released in February, Sonnet 5 shows marked improvements in reasoning, tool use, coding, and knowledge tasks. While Opus 4.8 remains the preferred choice for higher accuracy, Sonnet 5 offers developers a cost-effective alternative with much better quality than earlier options. Users can balance cost and performance by choosing between Sonnet 5 and Opus 4.8.

Independent experts cited in Anthropic's blog praised Sonnet 5 for completing complex multi-step tasks without interruption, a limitation in earlier versions. Senior engineer Daniel Shepard from Zapier highlighted the model's ability to autonomously update Salesforce account levels and send launch notifications, tasks that previously stalled mid-process. The model also demonstrates enhanced safety, with fewer undesirable behaviors such as misuse facilitation, hallucinations, and sycophantic responses compared to Sonnet 4.6, though it still trails Opus 4.8 and Claude Mythos in alignment and safety metrics.

This launch follows recent U.S. government restrictions on distributing Anthropic's most powerful models, Claude Mythos and Claude Fable, due to security concerns, especially in cybersecurity capabilities. Recently, the government authorized Mythos's release to over 100 U.S. organizations, reflecting ongoing regulatory scrutiny of advanced AI models.