After DeepSeek V4 Flash launched on Aug. 1, its daily token consumption on OpenCode rose from about 3 trillion to 18 trillion within two weeks, a sixfold increase that approached the entire OpenRouter platform’s daily processing volume. The model costs about one-350th as much as Fable, one-175th as much as GPT-5.6 Sol and one-70th as much as Claude Sonnet, leading users to describe it as AI that is “so cheap it doesn’t need to be measured.” OpenCode raised peak-hour prices fivefold and off-peak prices 2.5-fold on Aug. 16, after which daily usage fell more than 50% from its peak. Jay V said the increase was a passive response to rising GPU costs. OpenCode is seeking suppliers able to handle current traffic at the original prices, requiring about 1,000 B300 units. The episode highlights an “impossible triangle” in the AI model market: low prices, strong performance and stable large-scale supply are difficult to achieve simultaneously. OpenCode Go cut its quota from $60 to $15 before restoring it to $30 and introduced peak- and off-peak rates.