Anthropic Launches Claude Sonnet 5, Optimized for Agentic Tasks

Full analysis

Anthropic has released Claude Sonnet 5, positioning it as its most capable mid-tier model for agentic workloads — tasks where an AI model autonomously plans, uses tools (like browsers and terminals), and executes multi-step workflows without constant human guidance. The model narrows the performance gap with the flagship Opus 4.8 class while launching at introductory pricing of $2 per million input tokens and $10 per million output tokens (through August 31, 2026), after which prices rise to $3/$15 per million tokens. An updated tokenizer means the same input may use 1.0–1.35× more tokens than Sonnet 4.6, though Anthropic says introductory pricing is calibrated to make the transition roughly cost-neutral.

Early-access partners report the model completes complex coding, debugging, and multi-step automation tasks — such as updating a CRM and sending a launch email in a single uninterrupted run — where prior Sonnet models would stall. Safety evaluations show lower rates of hallucination, sycophancy, and susceptibility to prompt-injection attacks compared to Sonnet 4.6, though it scores slightly higher on misaligned-behavior tests than Opus 4.8. Anthropic deliberately avoided cybersecurity training; the model cannot produce working browser exploits, and cyber safeguards (real-time detection and blocking) are enabled by default. Sonnet 5 is available now across Free, Pro, Max, Team, and Enterprise Claude plans, in Claude Code, and via the Claude API as claude-sonnet-5.