Anthropic Targets Cost Efficiency with Claude Sonnet 5 Launch

The model represents a significant shift in the competitive landscape, where the primary differentiator among AI labs is no longer just performance, but the ability to deliver agentic work cheaply and without constant human oversight. Starting Tuesday, Sonnet 5 replaces previous versions as the default model for all Claude subscription tiers. Pricing is set at $2 per million input tokens and $10 per million output tokens through August 31, before increasing to $3 per million input tokens. This places the model below the cost of its predecessor, Opus 4.8, as well as competing offerings like OpenAI’s GPT-5.5.

Technically, the model shows measurable gains over the Sonnet 4.6 release from February, particularly in software coding and complex knowledge work. Anthropic notes that the model can now self-correct and verify its own output during multi-step processes, such as updating CRM records or managing enterprise communications. While Opus 4.8 remains the superior choice for high-stakes judgment calls, Sonnet 5 offers a high-utility alternative for day-to-day automation. Furthermore, the lab reports a reduction in undesirable behaviors, including improved resistance to prompt-injection attacks and lower rates of hallucination compared to earlier versions.