What's this about?
OrangeAI documents that cloud vendors reselling DeepSeek V4 Flash charge roughly 10× DeepSeek's own official API price — "as I predicted," he notes wryly.
What exactly is going on with model pricing here?
The same open/cheap model can reach you through the lab's own API or through a cloud reseller who hosts it and marks it up for convenience, SLAs, or captive enterprise buyers. The 10× gap shows "the price of a model" is really the price of *who you buy it from*.
Why does it matter?
85 likes. A concrete data point in the week's cost-transparency thread (Cursor efficiency, cheap Luna): the sticker price of open models can be wildly inflated downstream, which matters for anyone budgeting production inference.