DeepSeek pricing fuels loss-leader debate

This is speculative, but it is exactly the kind of market analysis AI developers should watch because pricing shifts can change model adoption faster than benchmark charts do. Even if the post overstates the subsidy angle, it surfaces a real question about how sustainable ultra-cheap inference really is.

• –DeepSeek’s official API docs do show unusually aggressive pricing and explicit support for context caching, so the broader concern about low prices is grounded in real product behavior.
• –The core argument is that shared prompt caching can explain cheap repeated workloads, but not fully explain why random cache-miss requests are still priced so low.
• –Comparing DeepSeek with Qwen 3.5 makes the thread more useful than generic pricing complaints because it ties business strategy to serving-architecture tradeoffs.
• –The practical takeaway for developers is simple: cheap APIs are great for prototyping and scale, but they are risky assumptions if the pricing is strategic rather than structural.