Open-weight models process most AI tokens, but capture a quarter of spend
Open-weight models now handle more than seven in ten AI Gateway tokens, giving teams cheaper capacity while taking nearly one in four dollars. Their spend share has more than quadrupled since June, even as September gateway spending reached about 13 times December's level. The shift comes from cheap, high-volume models such as DeepSeek-v4.1-Flash, which processed nearly half of September's tokens but captured just 2.6% of spend.