both frontier labs cut prices on the same afternoon. anthropic shipped claude opus 5.5 at $4 per million input tokens and $20 per million output, 20% below opus 5 and 60% cheaper on cache reads, the line item that dominates agentic work. ninety minutes later openai answered with gpt-6 sol at $2 and $10, and luna at $0.10 and $0.50, half their previous promo rates. opus 5.5 performs near fable 5.1 on most tasks. sol and luna hold most of the flagship capability at a fraction of the flagship price. the benchmark race is over. the competition moved to cost per task. when a model answers one prompt, pricing is a detail on a pricing page. when an agent works for hours, cost per successful task becomes an engineering constraint. openai says its median researcher now burns $600 a day in tokens. the 90th percentile burns $7,000. the moat moved from the model to the system around it: routing, evals, proprietary data, the product loop that converts raw intelligence into reliable outcomes. intelligence went from a badge to a budget line.