Anthropic launches Claude Sonnet 5 for cheaper, more autonomous AI workflows

Anthropic has introduced Claude Sonnet 5, positioning it as the company’s most capable Sonnet model for autonomous, tool-using work. The model can plan tasks, interact with browsers and terminals, and operate through extended workflows that previously demanded a more expensive Opus model.

For AI builders, the main change is the cost-performance tradeoff. Anthropic says Sonnet 5 approaches the capabilities of Opus 4.8 in some agentic evaluations, while improving substantially over Sonnet 4.6 in reasoning, coding, tool use, and knowledge work. Developers can also adjust the model’s effort level to balance speed, capability, and spending—particularly useful for agents that perform searches, manipulate computers, or complete multi-step software tasks.

The API model is available as claude-sonnet-5 at $2 per million input tokens and $10 per million output tokens. It is also the default for Claude’s Free and Pro plans, with access for Max, Team, and Enterprise customers.

Anthropic says early users have reported stronger follow-through on complex coding, debugging, research, and business automation tasks. The company’s safety evaluations found fewer undesirable behaviors than in Sonnet 4.6 and described the model as generally safer for agentic use. However, it also reports that Sonnet 5 is substantially less capable at cybersecurity tasks than its current Opus models.

The launch gives developers another option between lightweight automation and premium reasoning. Teams building agents may be able to reserve Opus for especially demanding cases while using Sonnet 5 for larger volumes of routine, multi-step work.

Source: Anthropic News