Anthropic Makes Claude Sonnet 5 the Default Model for Free Users

Anthropic has released Claude Sonnet 5, a new mid-sized model that immediately became the default choice for users on the Free and Pro plans. The company positions it as a model that performs close to the flagship Opus 4.8 but costs noticeably less to run, a point that matters especially for businesses running agentic tasks at scale.

What the model brings

According to Anthropic, Sonnet 5 is the most agentic version of Sonnet to date. The model can independently plan multi-step tasks, use tools such as a browser or terminal, and carry out longer work sessions without stalling partway through, capabilities that previously required larger and more expensive models.

In agentic coding benchmarks, Sonnet 5 scored 63.2 percent, compared with 58.1 percent for its predecessor Sonnet 4.6 and 69.2 percent for Opus 4.8. On knowledge-work tasks, Anthropic's data shows the new model slightly outperforming even Opus 4.8. The company also highlights improvements in reasoning, tool use, and code debugging.

Pricing and availability

Until August 31, the model is available at an introductory price of $2 per million input tokens and $10 per million output tokens. After that date, prices rise to $3 and $15 per million tokens. That's still cheaper than Opus 4.8, OpenAI's GPT-5.5, or Google's Gemini 3.1 Pro, though more expensive than the lightweight Gemini 3.5 Flash.

The model is now the default for Free and Pro accounts, as well as for Max, Team, and Enterprise plans. Developers can call it through the API under the name claude-sonnet-5, and it's also integrated into Claude Code and the Claude platform.

Safety and limitations

Anthropic says Sonnet 5 shows a lower rate of undesirable behaviors than Sonnet 4.6, including fewer instances of deception, hallucination, and excessive deference to users. The model is also better at refusing malicious requests and more resistant to prompt injection attacks when operating in agentic mode.

At the same time, the company notes that Sonnet 5's cybersecurity capabilities are noticeably weaker than those of the Opus family, and its default safeguards are less restrictive than those of Fable 5, which was previously temporarily pulled from some markets for safety reasons.

What it means for the market and Polish companies

The launch fits a broader trend in which agentic capabilities are becoming standard across all major providers, from OpenAI's GPT-5.6 Sol to Google's Gemini 3.5 Flash. Competition is shifting from raw model capability toward price-to-performance ratio and reliability on long, autonomous tasks.

For Polish companies deploying process automation or coding agents, cheaper tokens without a quality trade-off translate into real savings on operating costs, particularly where models run long, multi-step tasks without human supervision. It's also a signal that companies should keep testing new model versions regularly, since price differences between providers can run to several dozen percent.

Sources: Introducing Claude Sonnet 5 (anthropic.com), Anthropic launches Claude Sonnet 5 as a cheaper way to run agents (techcrunch.com)