DeepSeek V4 Runs a Million Tokens for $0.87 — Then Adds Surge Pricing Every Afternoon

Editor's take

DeepSeek has introduced a new version of its large language model capable of processing one million tokens for under a dollar, while also implementing dynamic pricing that increases costs during peak usage hours.

This development is significant as it democratizes access to extremely long context windows, a capability previously exclusive to proprietary models like Anthropic's Claude 2.1 (which handles 200,000 tokens). DeepSeek's move, particularly the surge pricing, signals a potential shift in how open-weight model providers monetize their infrastructure and manage demand, impacting researchers and developers who rely on cost-effective access.

Future developments will focus on the sustainability of this pricing model. It will be crucial to observe if other open-weight providers adopt similar dynamic pricing strategies and how this affects the broader adoption of long-context models outside of large enterprises.