Claude Sonnet 5 and new models drive API prices lower

API pricing saw a new round of cuts in July. Claude Sonnet 5 launched at a promotional $2/$10 per million tokens — just 40% of Opus 4.8 pricing. DeepSeek V4 Flash maintains ultra-low pricing at $0.14/$0.28. GPT-5.6 Terra is priced around $2.50/$15 (GA pricing expected). Gemini 3.5 Flash is $1.50/$9. The result: frontier model inference costs have dropped 70-85% compared to 2024, making intelligent routing and caching strategies key to cost control.