DeepSeek V4 Pro now runs updated weights on AI Gateway

Vercel’s AI Gateway has rolled out updated weights for DeepSeek V4 Pro, which are now the default for the existing model ID deepseek/deepseek-v4-pro. Users do not need to modify their code or change the model identifier to access the new weights. The update applies automatically to all requests made to the gateway using the standard model ID. Vercel states that this change ensures users benefit from the latest improvements without altering their integration setup.

To explicitly use the updated weights, developers must specify the new model identifier deepseek/deepseek-v4-pro-0813 in their requests. This identifier directs AI Gateway to route traffic to providers hosting the refreshed weights. Vercel emphasizes that the transition is seamless, requiring only a minor adjustment in model configuration for those who choose to opt in explicitly.

For users running DeepSeek V4 Pro within coding agents, Vercel provides a dedicated setup through AI Gateway. The process involves connecting agents to the gateway and selecting the new model identifier deepseek/deepseek-v4-pro-0813 in the agent’s configuration. Vercel offers a coding agents guide to assist with the integration, ensuring compatibility with the updated weights.

AI Gateway maintains its pricing model, reflecting provider costs without additional markup or platform fees on inference. This applies to both standard requests and Bring Your Own Key (BYOK) scenarios. Users can explore further details, review the AI Gateway model leaderboard, or test the service in the model playground provided by Vercel.