According to the Beating Monitor, Vercel released the AI Gateway Production Index for June 2026. The report reveals that following the launch in May, the DeepSeek V4 series of Vercel's gateway (including Flash and Pro models) saw a significant surge in Token traffic share, skyrocketing from less than 1% to 17% within a month, surpassing OpenAI (13%) for third place. However, due to its extremely low pricing, the total cost of all users deploying DeepSeek accounts for only around 1% of the gateway's overall expenditure.
Pricing has been the key driver behind DeepSeek's rapid growth. The cost for a million Token inputs and outputs with DeepSeek V4 Flash is only $0.14 and $0.28, making it 20 to 50 times cheaper than similar cutting-edge models from Anthropic, and 8 to 12 times lower compared to Qwen 3.6 Plus and Kimi K2.6. Performance evaluations have shown that DeepSeek V4 meets the standards, prompting development teams to swiftly deploy it in production.
Despite the surge in traffic from low-cost models, the expenditure is still dominated by cutting-edge models. In May, Anthropic's spending share increased from 61% to 65%, with 70% to 80% of the expenditure in high-complexity scenarios such as application generation, backend AI agents, and programming. For instance, in the programming AI agent scenario, DeepSeek contributed 49% of the Token traffic but accounted for only 4% of the costs, while Anthropic, with 28% of the traffic, consumed 70% of the funds.
Development teams are managing budgets through intelligent routing, diverting high-frequency, low-risk tasks to low-cost models and only using cutting-edge models at critical junctures. Considerations for return on investment (ROI) have also slowed down model upgrades. For example, Google's launch of Gemini 3.5 Flash in May with higher pricing than the 3.0 version has resulted in a slow migration, with the 3.0 Flash still representing 90% of the Flash series traffic by the end of the month, while the 3.5 Flash only holds 7%. Meanwhile, AI agents have exhibited a high Token consumption density, with a quarter of the request volume consuming over half of the Tokens.
