Anthropic Introduces Claude Sonnet 5: Fast and Affordable AI
Anthropic, one of the leading companies in the field of artificial intelligence, has officially announced its new Claude Sonnet 5 language model. This update is the most balanced representative of the Claude family, significantly approaching the capabilities of the flagship Claude Opus 4.8 in performing complex tasks. The main feature of the new model is that it remains affordable while maintaining high performance. This is reported by Ixbt.com news provides.
Developers note that Claude Sonnet 5 is primarily optimized for "agentic scenarios" — systems capable of independently performing a sequence of actions. This model not only works with text but is also capable of independently solving complex multi-step tasks using a browser, terminal, and various API interfaces. According to ixbt.com, the new model has advanced significantly over previous generations in terms of planning and working with external tools.
Capabilities and Technical Specifications
Claude Sonnet 5 has shown significant growth in four main areas: reasoning, programming, use of external tools, and performing agentic tasks. Internal tests conducted by Anthropic show that in some cases, the new model achieves results on par with Claude Opus 4.8, which is several times more expensive. In particular, the model's efficiency in operating within a real computer environment was highly rated in complex benchmarks such as BrowseComp and OSWorld-Verified.According to early users, Claude Sonnet 5 has become much more stable in completing long-running processes. While the previous Sonnet 4.6 version sometimes left complex tasks unfinished, the new generation is capable of independently verifying its results and correcting errors without user instruction. This significantly simplifies the workflow for developers and data analysts.
Security and Terms of Use
Security remains a priority for Anthropic. Claude Sonnet 5 has become more resilient in rejecting harmful requests and resisting "prompt injection" attacks (commands aimed at deceiving the model). Additionally, hallucinations — cases where false information is presented as fact — have been reduced. It should be noted that the model is not specifically trained to organize cybersecurity attacks, and its capabilities in this regard are limited.Claude Sonnet 5 is currently open to all users, including those on Free and Pro tariff plans. It is also available via Claude Code and the API platform. Regarding pricing, the company has set the following rates:
- Per million input tokens — $2;
- Per million output tokens — $10;
- These promotional prices are valid until August 31, 2026.
Comments 0
…