https://www.anthropic.com/news/claude-sonnet-5
Gemini summarized this.
Anthropic's newly released Claude Sonnet 5 (Claude Sonnet 5) is a model focused on agentic functionality, which allows it to independently plan and utilize tools like browsers or terminals.
Compared to the previous Sonnet 4.6, agent performance has significantly improved in terms of reasoning, coding, and tool utilization. It achieves performance close to the top-tier model Opus 4.8 while being much more cost-effective.
Developers who participated in initial testing evaluated Sonnet 5 as excelling at completing complex multi-step software engineering tasks, bug fixes, and real-time data analysis without interruption, self-verifying along the way.
Furthermore, overall safety has been improved compared to previous models. It exhibits enhanced resistance to malicious requests and prompt injection attacks. For potentially risky cybersecurity tasks, performance is limited, and a real-time blocking guardrail is applied by default. Claude Sonnet 5 is currently designated as the base model for both free and pro plans, making it accessible to all users. It is also available on platforms such as the Claude API and Claude code.
I'm trying it out to see for myself.
While I can't say it's better than Opus 4.8, it's definitely fast.
▶ Original source: https://www.anthropic.com/news/claude-sonnet-5
