Anthropic has released Claude Sonnet 5, a mid-tier agentic model priced at $2/$10 per million tokens (introductory rate until August 31, 2026), rising to $3/$15 afterward. It delivers near-flagship Opus 4.8 performance on reasoning, coding, and tool use at less than half the cost. The model is designed for long-running autonomous tasks like browser and terminal control. A new tokenizer means the same text can use up to 1.35x more tokens, so the effective cost savings are somewhat smaller than the headline rate suggests. On agentic coding benchmarks, Sonnet 5 scored 63.2% vs Opus 4.8's 69.2%. Safety improvements include better prompt-injection resistance and reduced hallucination, though it falls short of Opus 4.8 on safety metrics. Real-time cyber safeguards are enabled by default. The pricing strategy is aimed at retaining developers as enterprise AI costs have ballooned.
