Anthropic Launches Claude Sonnet 5 to Power Affordable AI Agents

Anthropic has officially released Claude Sonnet 5, a midsize model engineered specifically to bridge the gap between high-end reasoning and cost-effective autonomous execution. By prioritizing "agentic" capabilities—the ability to use tools, plan, and iterate—Anthropic is positioning Sonnet 5 as the go-to engine for developers building automated workflows.

The Shift Toward Agentic Intelligence

In the current AI arms race, the industry is moving beyond simple chatbots toward autonomous agents. Anthropic’s release follows similar moves from OpenAI with GPT-5.6 Sol and Google with Gemini 3.5 Flash, signaling that agentic performance is the new baseline.

Claude Sonnet 5 is designed to act as an autonomous operator, capable of utilizing browsers and terminals to execute multi-step tasks. Unlike previous iterations that might stall during complex sequences, Sonnet 5 demonstrates a unique ability to "check its own output" and complete end-to-end workflows. For instance, Zapier engineers noted the model successfully completed a two-part task—updating Salesforce account tiers and sending enterprise launch announcements—a process that previously caused earlier models to fail mid-way.

Performance Benchmarks: Matching the Heavyweights

While Sonnet 5 is a midsize model, its performance metrics approach those of Anthropic’s flagship, Opus 4.8. On agentic coding benchmarks, Sonnet 5 scored 63.2%, significantly higher than its predecessor Sonnet 4.6 (58.1%), and trailing only slightly behind Opus 4.8 (69.2%).

Remarkably, in specific knowledge work benchmarks, Sonnet 5 actually outperformed Opus 4.8. This makes it a highly efficient choice for developers who need deep reasoning without the premium price tag of a top-tier model. Anthropic suggests that while Opus 4.8 remains the standard for extreme accuracy and subtle judgment, Sonnet 5 offers the ideal balance of quality and cost for day-to-day automation.

Aggressive Pricing and Safety Standards

To drive adoption, Anthropic has introduced a competitive pricing structure. Through August 31, Sonnet 5 is priced at $2 per million input tokens and $10 per million output tokens. Following this period, prices will adjust to $3 per million input tokens and $15 per million output tokens. This positioning makes Sonnet 5 more affordable than OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro, though it remains more expensive than Gemini 3.5 Flash.

Safety is equally critical in agentic deployments, where a model's ability to refuse malicious commands is paramount. Sonnet 5 shows a reduced rate of "undesirable behaviors," such as deception or cooperation with misuse, compared to Sonnet 4.6. It has also demonstrated improved resilience against prompt-injection attacks and a lower rate of sycophantic behavior, making it a more reliable partner for builders deploying tools to millions of users.

Key Takeaways

  • Agentic Focus: Sonnet 5 is optimized for autonomous tasks, including tool use (browsers/terminals) and self-correction, making it ideal for complex automation.
  • Cost-Efficiency: The model offers a high-performance, lower-cost alternative to flagship models like Opus 4.8, GPT-5.5, and Gemini 3.1 Pro.
  • Enhanced Safety: Significant improvements in refusing malicious requests and resisting prompt injections make it safer for agentic workflows.