ANTHROPIC launched the Claude Sonnet 5 model, calling it the most capable Sonnet for agent tasks to date. The firm says Sonnet 5 can plan and invoke tools such as browsers and terminals with autonomous performance that previously required larger, costlier models. Performance approaches Opus 4.8 while offering a lower price; Sonnet 5 is materially stronger than Sonnet 4.6 on inference, tool use, programming and knowledge-work tasks. Internal safety tests show a lower incidence of undesirable beha

ANTHROPIC launched the Claude Sonnet 5 model, calling it the most capable Sonnet for agent tasks to date. The firm says Sonnet 5 can plan and invoke tools such as browsers and terminals with autonomous performance that previously required larger, costlier models. Performance approaches Opus 4.8 while offering a lower price; Sonnet 5 is materially stronger than Sonnet 4.6 on inference, tool use, programming and knowledge-work tasks. Internal safety tests show a lower incidence of undesirable behavior versus Sonnet 4.6, and Sonnet 5 is significantly weaker than current Opus models on cybersecurity tasks. The model is live across all plans — default on Free and Pro, opt-in for Max, Team and Enterprise — and is available on Claude Code and the Claude platform. Launch pricing is $2 per MLN input tokens and $10 per MLN output tokens through Aug. 31, 2026, rising to $3/$15 thereafter.