Anthropic's Claude Sonnet 5 matches Opus 4.8 at half the price
Sonnet 5 narrows the gap to Opus with superior agentic capabilities and lower cost.
Anthropic unveiled Claude Sonnet 5, a major leap in agentic AI for the mid-tier Sonnet line. Built to handle complex autonomous tasks like planning, browser control, and terminal use, Sonnet 5 achieves performance close to Opus 4.8 — a model that costs nearly twice as much ($5/MTok input, $25/MTok output). On agentic evaluations like BrowseComp and OSWorld-Verified, Sonnet 5 shows strict improvements over Sonnet 4.6, with higher-effort settings matching Opus on some benchmarks. Introductory pricing ($2 input, $10 output per million tokens) runs through August 31, 2026, making it highly cost-effective for scaling AI agents.
Safety assessments reveal Sonnet 5 has a lower rate of undesirable behaviors than Sonnet 4.6 and is generally safer in agentic contexts. Notably, it shows a much lower ability to perform cybersecurity tasks than current Opus models, reducing risk in automated workflows. The model is now the default on Free and Pro plans, and available to Max, Team, and Enterprise users. It also integrates with Claude Code and the Claude API, letting developers fine-tune effort levels to balance cost and performance. Early testers report it finishing multi-step tasks that stalled on earlier Sonnets, such as updating Salesforce records and sending launch emails end-to-end.
- Introductory pricing at $2/MTok input and $10/MTok output through August 31, then $3/$15 – half the cost of Opus 4.8
- Matches or approaches Opus 4.8 on agentic benchmarks (BrowseComp, OSWorld-Verified) with higher effort settings
- Improved safety with lower undesirable behavior rates and reduced cybersecurity capabilities compared to Opus
Why It Matters
Sonnet 5 democratizes advanced agentic AI, enabling complex automation at a fraction of Opus cost.