Open-weight models reached 56% of total token volume on Vercel AI Gateway in August 2026, increasing from 11% in April. On August 22, the share reached 62%, with recent data showing peaks at 78.4%.
Despite the volume shift, Anthropic maintains 61-65% of total gateway spending while processing 30% of tokens. Its models cost up to 4.4 times the average token price, reflecting a preference for premium quality in high-stakes tasks.
DeepSeek contributes approximately 25% or more of total token share, frequently leading in volume. Google accounts for 11% of the volume. Open-weight models from labs like DeepSeek, Z.ai, and Moonshot are used for high-volume, low-cost tasks.
The average price per token decreased by 23.2% in August, marking three consecutive months of decline. Over the past five months, prices have fallen by more than 50%. Open-weight models currently cost about one-seventh of frontier closed models.
