Silicon Valley’s AI pricing competition intensified over two days, with three companies releasing four models, including new offerings from GPT and Claude, pushing prices toward DeepSeek’s territory. Yet lower listed rates may not translate into lower overall costs. In a developer test, Opus 5.5 at its highest thinking tier took nearly 20 minutes and cost $2.56, but still failed to produce an image of a pelican riding a bicycle. The test highlights how expenses can extend beyond the base model price: extended reasoning and retries are billed, while human review also consumes time.