🧠 Food for thought

Implications, context, and why it matters.

DeepSeek’s business model for the new V4 models is usage-based and flexible

• DeepSeek sells access to the V4 models through a usage-based API. Customers pay for what they use instead of a fixed subscription.
• Charges are set per 1 million input and output tokens. DeepSeek’s pricing page says billing depends on total tokens processed, with no fixed monthly plan listed 1.
• Payments come from a topped-up balance or a granted balance. If both are available, the granted balance is used first 1.
• DeepSeek says prices may change, though it does not describe current rates as introductory offers 1.
• The company is also narrowing its lineup. It says deepseek-chat and deepseek-reasoner will be deprecated, and that they map to the non-thinking and thinking modes of deepseek-v4-flash 1.

DeepSeek’s pricing strategy could commoditize the AI model market

• A clear pay-per-use approach raises competition on price and performance, instead of model specs alone.
• That gives software developers and business buyers a simpler way to compare the cost of completing a task across providers.
• Those side-by-side checks could make AI models feel more interchangeable, which may lead more businesses to switch vendors based on price.
• Lower costs may also make some AI uses practical, especially work that needs large context windows. That pressure could push established providers to revisit bundled and tiered pricing.