DeepSeek 'Cuts Prices' by 75%, Except They Never Actually Cost That Much

DeepSeek just announced that its V4-Pro API is officially 75% cheaper for the long haul. For anyone paying attention, the price for a million input tokens is now $0.435, down from the theoretical $1.74, while output tokens dropped to $0.87 from a steep $3.48. It sounds like a massive holiday gift for developers, but in reality, this is just a formal confirmation of the promo rates everyone has been using since late April.

This pricing strategy is essentially a masterclass in marketing optics. The initial 'official' price list had the lifespan of a mayfly, existing only for a few days before a rolling series of promotions effectively replaced it. By locking these rates in permanently, the company is finally providing some budget predictability rather than an actual discount.

The competitive landscape is getting increasingly lopsided. While domestic rivals like Kimi and Zhipu are busy hiking their prices to chase margins, DeepSeek is doubling down on the aggressive, near-loss-leader strategy that started with their R1 model. Most notably, the cache-hit cost is now sitting at a trivial $0.003625 per million tokens, making long-term agentic workflows virtually free.

This is the kind of corporate gaslighting that actually benefits the end user, but it raises a massive question about the sustainability of the AI bubble. When one player is setting prices that are 140 times cheaper than the industry standard, it suggests either a bottomless pit of funding or that everyone else is just wildly overcharging for the same silicon compute.