Anthropic Releases Claude Sonnet 5 With Enhanced Agentic Capabilities

Anthropic released Claude Sonnet 5 this week, positioning it as their most agentic Sonnet model to date. This new iteration is designed to plan, operate browsers and terminals, and execute tasks autonomously over extended periods. Sonnet 5 is now the default model for Anthropic's Free and Pro subscription plans, and is also available to Max, Team, and Enterprise users. It is integrated into Claude Code and accessible via the Claude Platform.

According to Anthropic, Sonnet 5 significantly narrows the performance gap with their flagship Opus 4.8 model, particularly in agentic tasks. On key benchmarks, Sonnet 5 achieved 63.2% on SWE-bench Pro, 81.2% on OSWorld-Verified, and 57.4% on HLE, surpassing its predecessor, Sonnet 4.6, across all metrics. The model also introduces more competitive API pricing, with introductory rates of $2 per million tokens for input and $10 per million for output through August 31, before increasing to $3 and $15 respectively. This pricing structure makes it a cost-effective option for low to medium effort tasks, though for extra-high effort tasks, the cost may approach or exceed that of Opus 4.8 for comparable quality.

Sonnet 5 is positioned between the more affordable Haiku 4.5 and the high-performance Opus 4.8. This release emphasizes agentic reliability and improved context retention for longer task chains, along with enhanced self-correction capabilities when tool calls fail. Anthropic also noted that Sonnet 5 exhibits deliberately lower cyber capabilities compared to Sonnet 4.6, making Opus the preferred choice for accuracy-critical applications. The model utilizes an updated tokenizer, consistent with Opus 4.7, which may result in a higher token count for the same text, ranging from 1.0 to 1.35 times more tokens.