Anthropic Unveils New Model 「Claude Sonnet 5」

Anthropic has released a new AI model 「Claude Sonnet 5」. Positioned in the mid-tier of the company's lineup, this model is designed to deliver performance approaching its top-tier offering while being available at substantially lower cost. In both pricing structure and performance metrics, it is positioned as narrowing the gap with higher-end models.

As competition in the AI model space intensifies, companies are increasingly pursuing strategies to simultaneously achieve 'high performance' and 'low cost'. Anthropic itself is at a stage of preparing for its IPO, with the aim of expanding its user base by making the technology available to a broad range of developers and enterprises. The release of Sonnet 5 can be viewed as part of this strategic direction.

On the pricing front, the API adoption period (through August 31st) is set at 2 dollars per 1 million input tokens and 10 dollars per 1 million output tokens. After this period, pricing will transition to 3 dollars for input and 15 dollars for output, but this still falls significantly below the top-tier model 「Opus 4.8」at 5 dollars for input and 25 dollars for output. The model will serve as the default for Free and Pro plans, while also being available on Max, Team, and Enterprise plans.

On the performance front, multiple evaluation metrics show that Sonnet 5 achieves substantial improvements over its predecessor 「Sonnet 4.6」and approaches Opus 4.8 performance levels. On 「SWE-bench Pro」, which measures AI coding capabilities, Sonnet 5 recorded 63.2% compared to Sonnet 4.6's 58.1%, narrowing the gap with Opus 4.8's 69.2%. On an alternative coding evaluation, 「Terminal-Bench 2.1」, Sonnet 5 achieved 80.4%, nearly matching Opus 4.8's 82.7%.

On 「Humanity's Last Exam」, which measures reasoning ability across diverse domains, Sonnet 5 recorded 57.4% under tool-use conditions, nearly equivalent to Opus 4.8's 57.9%. Additionally, on the knowledge-work benchmark 「GDPval-AA v2」, Sonnet 5 achieved a score of 1,618, marginally exceeding Opus 4.8's 1,615. These results demonstrate that at approximately 60% lower standard pricing, the model reaches performance levels comparable to the flagship model.

What Anthropic particularly emphasized in this release is the capability for 'agentic AI'. Agentic AI refers to systems that not only answer questions but can autonomously execute multi-step tasks while utilizing tools such as browsers and terminals. As of 2026, the role enterprises seek from AI is shifting from 'conversational' to 'autonomous task execution', and Sonnet 5 is positioned as a model addressing this demand.

The release of Sonnet 5 can be interpreted as an effort to democratize high-performance AI, expanding it from 'something only for select large enterprises' to 'something accessible to cost-conscious developers'. Given Anthropic's business expansion and IPO preparation context, future points of attention will be the growth in user numbers and the extent to which agentic AI penetrates actual business operations.