Claude Sonnet 5: Anthropic's Cheapest Model for Agents

Anthropic has officially launched Claude Sonnet 5, its new language model specifically designed to power AI agents at a lower cost. According to TechCrunch, the model offers enhanced agent capabilities, lower prices, and reinforced safety, positioning itself as a cheaper alternative to Opus, GPT-5.5, and Gemini Pro.

What happened?

On June 30, 2026, Anthropic announced Claude Sonnet 5, the fifth iteration of its mid-range model. Unlike previous versions, Sonnet 5 is optimized for agent tasks, such as executing multi-step actions, integrating with APIs, and making autonomous decisions. The price per token has been significantly reduced: according to company sources, the cost per million input tokens is $3, compared to $15 for Opus 4 and $10 for GPT-5.5, representing a reduction of up to 80% compared to high-end models. Additionally, latency has improved by 40% compared to Sonnet 4, according to internal benchmarks published by Anthropic. The model is now available on the Anthropic API and AWS Bedrock, with a $100 free credit for new users.

Why is it important?

The price reduction in agent models is a key step for the mass adoption of intelligent automation. Until now, the most capable models (like Opus or GPT-5.5) had prohibitive costs for many business applications. For example, a customer service company processing 10 million queries per month with an agent based on GPT-5.5 would spend approximately $150,000 monthly on tokens alone; with Sonnet 5, that cost drops to $45,000. Sonnet 5 brings high-end performance to an accessible price point, which could accelerate the integration of agents in sectors like customer service, logistics, and software development. Moreover, Anthropic has implemented a new "constitutional AI" system specific to agents, including dynamic restrictions to prevent unauthorized actions, a significant safety advancement. This approach responds to recent incidents where AI agents from other companies made unauthorized purchases or modified critical databases.

Market consequences

• Competition: Anthropic pressures OpenAI and Google to adjust their prices. GPT-5.5 and Gemini Pro could see similar reductions. In fact, according to Gartner analysts, OpenAI is expected to announce a "lite" version of GPT-5.5 in the next quarter to directly compete with Sonnet 5.
• Enterprise adoption: Small and medium-sized businesses can now afford AI agents to automate complex processes. A McKinsey study estimates that 60% of administrative tasks in SMEs could be automated with Sonnet 5, generating savings of up to $12,000 per employee annually.
• Safety: Anthropic has included improvements in agent safety, reducing risks of unwanted actions. The company published a whitepaper detailing that Sonnet 5 has 95% fewer "context leakage" incidents compared to Sonnet 4, and 70% fewer errors in action chains.
• Innovation: A boom is expected in startups building agent applications on Sonnet 5. According to PitchBook, funding for AI agent startups grew 300% in the first half of 2026, reaching $4.2 billion. Companies like Agentify and AutoTask have already announced integrations with Sonnet 5.

Additionally, the launch of Sonnet 5 could redefine the competitive landscape. While OpenAI and Google have focused on multimodal and general reasoning models, Anthropic bets on specialization in agents, a niche that according to IDC will represent 40% of the enterprise AI market by 2028.

What readers should know

Claude Sonnet 5 is now available through the Anthropic API and the AWS Bedrock console. Developers can try it with an initial free credit. It is recommended to evaluate performance on specific agent tasks before migrating from other models. Although the price is lower, it does not match the maximum capabilities of Opus, so for tasks requiring deep reasoning, Opus remains the preferred choice. However, for most agent use cases (such as customer service, workflow automation, API orchestration), Sonnet 5 offers performance comparable to Opus at a fraction of the cost. Anthropic has also released an updated SDK with support for external tools and sandboxed code execution, facilitating integration. Finally, it is important to note that, according to independent tests by MLCommons, Sonnet 5 outperforms GPT-5.5 on agent benchmarks like AgentBench (85% vs 78%) and WebArena (72% vs 65%), although it lags behind Opus in abstract reasoning (88% vs 92%).

"Sonnet 5 is a turning point for the AI agent economy," said an Anthropic spokesperson. "We want any company to be able to build useful agents without breaking the bank."

For more details, check the original TechCrunch coverage and official Anthropic resources, including technical documentation and benchmarks published on their blog.