On June 30, 2026, Anthropic released a new AI model—Claude Sonnet 5—described as the most autonomous ("agentic") version of the Sonnet lineup to date. The model significantly outperforms its predecessor, Sonnet 4.6, in tasks involving reasoning, tool use, programming, and knowledge-based work.
What's new
Agency and autonomy. The model has learned to plan actions better and use tools—such as the browser and terminal—more effectively, enabling it to execute complex, multi-step tasks with virtually no human intervention. According to early testers, it successfully completes tasks where previous versions would stall halfway through.
Programming. The model demonstrates strong performance across all stages of development—from planning and implementation to debugging, refactoring large-scale projects, and ongoing code maintenance.
Reasoning and knowledge work. Analytical capabilities, the quality of tool usage, and results in professional tasks—such as research, analysis, and business processes—have improved significantly.
Safety. According to Anthropic, Sonnet 5 exhibits lower levels of undesirable behavior compared to Sonnet 4.6 and is generally safer in agentic scenarios, although its capabilities in the realm of cybersecurity are noticeably lower than those of Opus-class models.
Availability and price
Since its release, Claude Sonnet 5 has been available across all tiers: it has become the default model for Free and Pro plans and is also available to Max, Team, and Enterprise users. The model also operates within Claude Code and on the Claude Platform, where an introductory price applies at launch: $2 per million input tokens and $10 per million output tokens through August 31, 2026, after which the price will rise to $3 and $15, respectively.
According to independent reviewers, Sonnet 5 offers performance close to that of the higher-end Opus 4.8 model, but at a significantly lower price—cheaper than Opus 4.8, OpenAI's GPT-5.5, and Google's Gemini 3.1 Pro.
Source: Anthropic — anthropic.com/news/claude-sonnet-5
0 Comments